perf: parallelize module-graph parsing

[ivov]

Parallelizes the parsing of every module's .lis files, draining one BFS-level at a time from a queue. Also stops registering each parsed file in the store twice.

lis build cold:

Modules	Before	After	Delta
1	56 ms	55 ms	-1%
10	89 ms	84 ms	-5%
30	159 ms	140 ms	-12%
100	507 ms	456 ms	-10%

lis build warm:

Modules	Before	After	Delta
1	48 ms	48 ms	0%
10	62 ms	50 ms	-20%
30	115 ms	83 ms	-28%
100	255 ms	209 ms	-18%

n modules ~660 LOC each, LISETTE_NO_CACHE=1, hyperfine with --warmup 5 --runs 25, M1 MBP

[github-actions]

Released in lisette-v0.2.3