Enable Python warnings during tests and fix all in the code

[jdufresne]

When running the test suite with Python warnings enabled, several
warnings are emitted. They look like:

tests/test_utils.py::test_format_requirement
  /usr/lib64/python3.9/tempfile.py:817: ResourceWarning: Implicitly cleaning up <TemporaryDirectory '/tmp/tmpezxehep1source'>
    _warnings.warn(warn_message, ResourceWarning)

To solve, make Repository objects a context manager. Upon entering the
with statement, the build an source caches are created and upon exit
they are cleaned up. This ensures they only live for the duration they
are required. This could theoretically be used creating and destroying
any necessary resources. Tests were adjusted.

tests/test_repository_pypi.py::test_pypirepo_build_dir_is_str
  .../pip/_vendor/requests/adapters.py:59: ResourceWarning: unclosed file <_io.FileIO name='/tmp/pytest-of-jon/pytest-70/test_open_local_or_remote_file5/foo.txt' mode='rb' closefd=True>
    super(BaseAdapter, self).__init__()

To solve, test_open_local_or_remote_file__remote_file() was adjusted to
close its file after use.

tests/test_cli_compile.py::test_redacted_urls_in_verbose_output[--find-links]
  .../pip-tools/piptools/scripts/compile.py:306: FutureWarning: --index and --no-index are deprecated and will be removed in future versions. Use --emit-index-url/--no-emit-index-url instead.
    warnings.warn(

To solve, use --no-emit-index-url in
test_redacted_urls_in_verbose_output().

There remains one warning left, but this is out of the control of
pip-tools. An upstream PR has been filed:
pypa/pip#9156

Enabling the warnings in test runs will help catch them earlier and make
debugging/fixing easier.

Changelog-friendly one-liner:

Contributor checklist

• Provided the tests for the changes.
• Gave a clear one-line description in the PR (that the maintainers can add to CHANGELOG.md on release).
• Assign the PR to an existing or new milestone for the target version (following Semantic Versioning).

[atugushev]

Since freshen_build_caches does nothing interesting but yield it can be declared as an abstract method (same as allow_all_wheels). What do you think?

[jdufresne]

it can be declared as an abstract method

Can you explain a bit how that helps us, what the benefit is, or what it works towards solving?

[jdufresne]

I went ahead with the abstract method approach. The reduction in repetition in indentation is nice.

One downside, this required adding an empty method to FakeRepository to fulfill the ABC requirements. This will be true of any new repository that doesn't require build caches. In practice, this probably won't be an issue going forward, just something to mention w/r/t the trade offs.

Thanks for the reviews!

[atugushev]

Thanks!