Add multi-process support to --check.

Set the parameter --num_threads to the desired number of worker tasks to
potentially speed up --check.

This works by spawning multiple sub-proccesses that each run the desired
diagnostics on a subset of the workspace. Each process will still load
and compile the entire workspace, so there are diminishing returns and
memory usage increases linearly with the number of threads.

Overall this can reduce the runtime by about ~50% for my projects,
example results:

Workspace 1, dominated by a few large/complex files

1 thread:  49.7 seconds
2 threads: 31.8 seconds
4 threads: 23.6 seconds
8 threads: 24.4 seconds

Workspace 2, large number of small-ish files

1 thread:  96.0 seconds
2 threads: 76.5 seconds
4 threads: 49.5 seconds
8 threads: 38.1 seconds

0 files changed, 0 insertions, 0 deletions