Add a command-line flag for overriding pthreadpool size (#46781)
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/46781
Test Plan: Passed it to speed_benchmark_torch and saw perf change.
Reviewed By: iseeyuan
Differential Revision: D24752889
Pulled By: dreiss
fbshipit-source-id: 762981510f271d20f76e33b6e6f361c4a6f48e6c