For image comparison tests in svg/pdf/ps formats, the result images are
converted to png for comparison. Previously the conversion results were
cached for the baseline images, but not for the test-generated images
(because of non-deterministic svg/pdf/etc. results, due to
hash-salting, dict ordering, etc.).

Now that the test-generated images are generally deterministic, we can
enable the cache for baseline images as well. This speeds up
pytest -k '[svg]' by ~30% (81s initially -> 55s on a seeded cache) and
pytest -k '[pdf]' by ~10% (62s -> 55s) (there are too few (e)ps image
comparison tests to see an effect). Also add logging regarding the
cache which may help troubleshooting determinacy problems.

This is a much simpler version of PR#7764, which added more sophisticated
reporting of cache hits and misses, as well as cache invalidation (I
think the cache can reasonably be cleaned manually, at least for now).

A simple cache eviction mechanism prevents the cache from growing
without bounds, limiting it to 2x the size of the baseline_images
directory.

This is a much simpler version of PR7764, which added more sophisticated
reporting of cache hits and misses and cache eviction.

Supersedes (mostly) #7764 (@jkseppan).

(Note that the gain in speed is less than reported than in #7764 (comment), because I have also optimized the svg and gs converters in the meantime by running a single process and interacting with it over stdin/stdout, rather that launching a new instance of inkscape/ghostscript for each image.)

PR Summary

PR Checklist

• Has Pytest style unit tests
• Code is Flake 8 compliant
• New features are documented, with examples if plot related
• Documentation is sphinx and numpydoc compliant
• Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
• Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way