memo parser: LRU capped cache

[benmosher]

turns out the memo parser really soaks up memory on large projects in v2.

setting: max number of modules to cache (50 or so by default?) to keep it lighter?

need to test cache hit frequency then, though.

first: measure perf improvement of memo-parser using v2, vs. not using it. maybe should just ditch it if it's not substantial anymore

[benmosher]

discovered while investigating #1058 (this explains why I see the same memory/perf in 2.9)

[benmosher]

another idea: use FS for cache instead of memory? needs perf observation also.

spec: some tmp folder with content-hash-based filenames holding AST JSON? actually maybe that's not achievable since the AST is not a DAG

[benmosher]

yeah, nvm, it doesn't seem to matter anymore so closing this