Create micro-benchmarks for core::Semaphore

[gavv]

Last revised: Oct 2023

It would be useful to measure Semaphore construction, destruction, post() when there are no waiters, post() when there are waiters, and non-blocking wait() (i.e. after post).

The numbers on Linux/Glibc are quite predictable (something compared to unlocked mutex without waiters, i.e. a few nanoseconds, and a minimum syscall overhead when there are waiters, i.e. a few hundred nanoseconds).

However, it would be interesting to get these numbers on uclibc (Linux), bionic (Android), and macOS (where core::Semaphore is implemented using kernel semaphores).

See #366 for details on adding micro-benchmarks.

[pranjii]

Can I get this?

[gavv]

@pranjii Sure!