Add retriableHttpCodes and reopenableExceptions so the user can, if they
choose, customize the conditions under which we retry an access or
reopen a channel after an exception closed it.

This can be handy if a cloud provider starts transiently throwing
exceptions that would normally indicate a fatal condition, but which in
practice will pass if retried.

The backoff logic in retries / reopens in unmodified.

Also add a few basic tests to make sure the retryable options do what we
expect.

In the built-in UNIX filesystem, the newDirectoryStream method
will return absolute paths if given absolute paths as input,
and relative paths if given relative paths as input.

We're now seeing this too for the NIO Cloud Storage provider
(this is a good thing), so I updated this test to reflect
this new reality.