PERF: fast paths for CRS equality and hashing

[greglucas]

Comparing two CRS objects goes through PROJ's proj_is_equivalent_to, which short-circuits on matching identifiers (EPSG codes) but otherwise falls through to a full recursive structural comparison of the datum, ellipsoid, prime meridian, coordinate system and conversion. For CRS built from PROJ strings, which carry no identifier, that costs 11-15us per comparison.

Noticed this in Cartopy where we do a check to see if self == src_crs to try and speed up transforms, but the CRS comparison itself can be slow.
https://github.com/SciTools/cartopy/blob/bdb6cc45e27ff4fe71741b445ac15449aee35ba3/lib/cartopy/crs.py#L410

The short circuits are the simple portion is self and srs1 == srs2, but the problem is that in missing cases we will still fall back to PROJ comparisons which is why I added the cache portion here which is really to speed up the != comparisons.

• Tests added
• Fully documented, including history.rst for all changes and api/*.rst for new API