Optimize sees by removing reachability cache and fixing LRU cache.

[prharvey]

Summary

Performance "Optimize creature vision checks."

Purpose of change

Creature vision checks are incredibly expensive, mostly due to the reachability cache thrashing.

Describe the solution

Removed the reachability cache entirely. It no longer serves a purpose now that reachability zones are used instead. Fixed a bug in the LRU cache implementation that made it an LRI cache instead. Rearrange some predicates in sees based on expense; note that this order will likely change again after my micro-optimizations PRs are merged.

Creature vision based on light level was also simplified. Looking from the shadows into light uses the max vision range of either light level, looking from light into shadows uses the min vision range of either light level. The current implementation makes no sense whatsoever, and looking at PR history doesn't make it clear at all. Removed and will PR separately.

Describe alternatives you've considered

Solve the cache thrashing problem in the reachability cache. Not worth the time given such map spanning caches are expensive to maintain anyways, especially if we are increasing reality bubble size, and it currently has limited utility. A less permissive version of this cache based on octants would be interesting, but it's just way too expensive to generate, currently.

Testing

Unit tests pass. Played for ~30 minutes and didn't notice anything off about monster vision.

Additional context

Note these flame graphs were taken using the new pathfinder, in order to make the change more obvious.

Flame before:

Flame after: