Fix getpelican#3037

Prevents FatalLogger from re-logging warning & error level log records,
which had resulted in masking the initial source of the log messages.

Change to Interface
===================
This change modifies the FatalLogger class attributes:
- warnings_fatal
- errors_fatal
+ fatal_log_level

Motivation
----------
Intention is to clarify the behavior: raise RuntimeErrors on any
desired log-level *or higher*.

The previous implementation did not respond to higher log levels (eg.
logging.CRITICAL) and allowed unintuitive states: for example, it was
possible to set the logger to fail at warning level yet not error-level.

Fix getpelican#3037

Overview
========
Motivated by issue getpelican#3037 (Ensure proper log line/filename in
warning/errors).

``pelican.log`` is no longer required to be early in the import order,
as the package no longer relies on a custom logging class.

Logging setup is now uncoupled so that format and filters can be applied
independently, which makes both setup and testing more flexible.

Behavior of commandline calls have not changed, except for some
modification to help texts.

``LOG_FATAL`` and ``LOG_ONCE_LEVEL`` are new configuration keys.

Summary of changes
==================

:pelican/__init__.py:

    * Parsing commandline:
        * Unify type-converstion from string to logging levels.
        * Tweak some help text in ``parse_arguments``.
    * Preserve ``log`` namespace throughout (for clarity).
    * FIX: ``pelican.listen`` verbosity: avoid assumption that the
      logger's level will be NOTSET.

:pelican/log.py:

    * FIX (getpelican#3037): warnings & errors now display the correct line numbers.
    * Split logging filters with focus on a specific aspect of behavior.
      This allows for more
        a) independence in testing and
        b) powerful logger configuration, with fewer side effects.
    * Isolate configuration *reading* from configuration *application*
      (see ``pelican/settings.py``).

:pelican/settings.py:

    * Remove ``pelican.log`` dependency.
    * Isolate configuration *reading* from configuration *application*;
      to remove dependency on a global state when modifying log filters.
    * Add validation for LOG_LIMIT values.

:pelican/tests/__init__.py:

    * Remove dependency on ``pelican.log`` module as test runners setup
      their own console handler anyway.
    * Only add ``NullHandler`` in absence of a configured logger.

:pelican/tests/support.py:

    * Remove ``LoggedTestCase`` class; tests requiring logging analysis
      use a contextmanager (``LogCountHandler.examine``) instead.

:pelican/tests/test_log.py:

    * Rewrite to reflect changes in pelican's logging module.
    * Increase code coverage.

:pelican/tests/test_contents.py:
:pelican/tests/test_pelican.py:
:pelican/tests/test_utils.py:

    * Import ``unittest`` directly, rather than from other modules.
    * Switch to using contextmanager (``LogCountHandler.examine``) for
      test-specific log inspection.
    * Remove dependency on a custom TestCase (used in logging analysis).
    * Remove code involved with temporarily disabling log filters --
      the filters in question are now only enabled in tests directly
      testing them (eg. in ``test_log.py``)

:pelican/tests/test_settings.py:

    * Add test for LOG_FILTER values from configs.

Overview
========
Motivated by issue getpelican#3037 (Ensure proper log line/filename in
warning/errors).

``pelican.log`` is no longer required to be early in the import order,
as the package no longer relies on a custom logging class.

Logging setup is now uncoupled so that format and filters can be applied
independently, which makes both setup and testing more flexible.

Behavior of commandline calls have not changed, except for some
modification to help texts.

``LOG_FATAL`` and ``LOG_ONCE_LEVEL`` are new configuration keys.

Summary of changes
==================

:pelican/__init__.py:

    * Parsing commandline:
        * Unify type-converstion from string to logging levels.
        * Tweak some help text in ``parse_arguments``.
    * Preserve ``log`` namespace throughout (for clarity).
    * FIX: ``pelican.listen`` verbosity: avoid assumption that the
      logger's level will be NOTSET.

:pelican/log.py:

    * FIX (getpelican#3037): warnings & errors now display the correct line numbers.
    * Split logging filters with focus on a specific aspect of behavior.
      This allows for more
        a) independence in testing and
        b) powerful logger configuration, with fewer side effects.
    * Isolate configuration *reading* from configuration *application*
      (see ``pelican/settings.py``).

:pelican/settings.py:

    * Remove ``pelican.log`` dependency.
    * Isolate configuration *reading* from configuration *application*;
      to remove dependency on a global state when modifying log filters.
    * Add validation for LOG_LIMIT values.

:pelican/tests/__init__.py:

    * Remove dependency on ``pelican.log`` module as test runners setup
      their own console handler anyway.
    * Only add ``NullHandler`` in absence of a configured logger.

:pelican/tests/support.py:

    * Remove ``LoggedTestCase`` class; tests requiring logging analysis
      use a contextmanager (``LogCountHandler.examine``) instead.

:pelican/tests/test_log.py:

    * Rewrite to reflect changes in pelican's logging module.
    * Increase code coverage.

:pelican/tests/test_contents.py:
:pelican/tests/test_pelican.py:
:pelican/tests/test_utils.py:

    * Import ``unittest`` directly, rather than from other modules.
    * Switch to using contextmanager (``LogCountHandler.examine``) for
      test-specific log inspection.
    * Remove dependency on a custom TestCase (used in logging analysis).
    * Remove code involved with temporarily disabling log filters --
      the filters in question are now only enabled in tests directly
      testing them (eg. in ``test_log.py``)

:pelican/tests/test_settings.py:

    * Add test for LOG_FILTER values from configs.

Overview
========
Motivated by issue getpelican#3037 (Ensure proper log line/filename in
warning/errors).

``pelican.log`` is no longer required to be early in the import order,
as the package no longer relies on a custom logging class.

Logging setup is now uncoupled so that format and filters can be applied
independently, which makes both setup and testing more flexible.

Behavior of commandline calls have not changed, except for some
modification to help texts.

``LOG_FATAL`` and ``LOG_ONCE_LEVEL`` are new configuration keys.

Summary of changes
==================

:pelican/__init__.py:

    * Parsing commandline:
        * Unify type-converstion from string to logging levels.
        * Tweak some help text in ``parse_arguments``.
    * Preserve ``log`` namespace throughout (for clarity).
    * FIX: ``pelican.listen`` verbosity: avoid assumption that the
      logger's level will be NOTSET.

:pelican/log.py:

    * FIX (getpelican#3037): warnings & errors now display the correct line numbers.
    * Split logging filters with focus on a specific aspect of behavior.
      This allows for more
        a) independence in testing and
        b) powerful logger configuration, with fewer side effects.
    * Isolate configuration *reading* from configuration *application*
      (see ``pelican/settings.py``).

:pelican/settings.py:

    * Remove ``pelican.log`` dependency.
    * Isolate configuration *reading* from configuration *application*;
      to remove dependency on a global state when modifying log filters.
    * Add validation for LOG_LIMIT values.

:pelican/tests/__init__.py:

    * Remove dependency on ``pelican.log`` module as test runners setup
      their own console handler anyway.
    * Only add ``NullHandler`` in absence of a configured logger.

:pelican/tests/support.py:

    * Remove ``LoggedTestCase`` class; tests requiring logging analysis
      use a contextmanager (``LogCountHandler.examine``) instead.

:pelican/tests/test_log.py:

    * Rewrite to reflect changes in pelican's logging module.
    * Increase code coverage.

:pelican/tests/test_contents.py:
:pelican/tests/test_pelican.py:
:pelican/tests/test_utils.py:

    * Import ``unittest`` directly, rather than from other modules.
    * Switch to using contextmanager (``LogCountHandler.examine``) for
      test-specific log inspection.
    * Remove dependency on a custom TestCase (used in logging analysis).
    * Remove code involved with temporarily disabling log filters --
      the filters in question are now only enabled in tests directly
      testing them (eg. in ``test_log.py``)

:pelican/tests/test_settings.py:

    * Add test for LOG_FILTER values from configs.

Overview
========
Motivated by issue getpelican#3037 (Ensure proper log line/filename in
warning/errors).

``pelican.log`` is no longer required to be early in the import order,
as the package no longer relies on a custom logging class.

Logging setup is now uncoupled so that format and filters can be applied
independently, which makes both setup and testing more flexible.

Behavior of commandline calls have not changed, except for some
modification to help texts.

``LOG_FATAL`` and ``LOG_ONCE_LEVEL`` are new configuration keys.

Summary of changes
==================

:pelican/__init__.py:

    * Parsing commandline:
        * Unify type-converstion from string to logging levels.
        * Tweak some help text in ``parse_arguments``.
    * Preserve ``log`` namespace throughout (for clarity).
    * FIX: ``pelican.listen`` verbosity: avoid assumption that the
      logger's level will be NOTSET.

:pelican/log.py:

    * FIX (getpelican#3037): warnings & errors now display the correct line numbers.
    * Split logging filters with focus on a specific aspect of behavior.
      This allows for more
        a) independence in testing and
        b) powerful logger configuration, with fewer side effects.
    * Isolate configuration *reading* from configuration *application*
      (see ``pelican/settings.py``).

:pelican/settings.py:

    * Remove ``pelican.log`` dependency.
    * Isolate configuration *reading* from configuration *application*;
      to remove dependency on a global state when modifying log filters.
    * Add validation for LOG_LIMIT values.

:pelican/tests/__init__.py:

    * Remove dependency on ``pelican.log`` module as test runners setup
      their own console handler anyway.
    * Only add ``NullHandler`` in absence of a configured logger.

:pelican/tests/support.py:

    * Remove ``LoggedTestCase`` class; tests requiring logging analysis
      use a contextmanager (``LogCountHandler.examine``) instead.

:pelican/tests/test_log.py:

    * Rewrite to reflect changes in pelican's logging module.
    * Increase code coverage.

:pelican/tests/test_contents.py:
:pelican/tests/test_pelican.py:
:pelican/tests/test_utils.py:

    * Import ``unittest`` directly, rather than from other modules.
    * Switch to using contextmanager (``LogCountHandler.examine``) for
      test-specific log inspection.
    * Remove dependency on a custom TestCase (used in logging analysis).
    * Remove code involved with temporarily disabling log filters --
      the filters in question are now only enabled in tests directly
      testing them (eg. in ``test_log.py``)

:pelican/tests/test_settings.py:

    * Add test for LOG_FILTER values from configs.