Feature or enhancement

The lockbench.py benchmarks PyMutex and PyThread_type_lock for various number of threads and length of critical section. You'll see similar benchmarks for other mutex implementations: Rust parking_lot, WebKit.

I'd like to make a few updates to this benchmark:

• Remove PyThread_type_lock. It's now based on PyMutex, so this isn't really useful anymore.
• Add --work-outside to adjust the amount work done while not holding the lock. This helps simulate low and moderate contention scenarios.
• Add --num-locks. So if we run with 8 threads and 2 locks, each group of 4 threads is independently contending for a lock. This adds more concurrency and helps demonstrate the limits of excessive spinning.
• Make number of threads configurable
• Add --total-iters which configures each thread to run for a fixed number of iterations instead of the default of running the whole benchmark for 1 second. With this mode, the benchmarks is only as fast as the slowest thread. It's another way to measure the effect of "fairness".

Linked PRs

• gh-145230: Update lockbench #145231
• gh-145230: Update lockbench.py and benchmark_locks API #145271