Load target build files in parallel using Python multiprocessing.
This parallelizes the portion of the processing that took the largest fraction of runtime previously. There's more opportunity for parallelization elsewhere, but this seems to have the biggest impact.
I did some testing to verify the output is identical, but mostly with ninja. Someone familiar with all of the generators should make sure I'm not making assumptions I shouldn't be. Also, it might make sense to keep this behind an optional flag or env var for now.
1. Linux, z600, chromium src + src-internal + full WebKit: 23.14 s ->
16.04 s (30% speedup)
2. MacBook Pro, chromium src only: 36.83 s -> 27.87 (25% speedup)
Total comments: 12
Total comments: 9
Total comments: 8
Total messages: 13