Implement basic pthread affinity support

[Menkib64]

This reserves an unique cores for SearchWorker. Assumption is that a second thread on the same core would increase latency when SearchWorker has a deadline to supply more data to GPU. It uses lscpu because it looks like the easiest way to get information about NUMA and threads sharing a core. There is a basic fallback implementation because lscpu is an optional utility.

Windows support is missing. I had a quick look into Windows API. I couldn't immediately figure out correct way to do the same thread binding.

[Menkib64]

meson configure on circleci Ubuntu fails to detect pthread_setaffinity_np. It does work when I test it locally. Search worker thread affinity settings require passing --threads=2 because I couldn't figure out a way to implement auto detection without creating backend threads. I try to figure out a solution how to change affinity for existing threads when we know thread number.

Good way to check thread affinities is to run

$ ./lc0 dag-preview -t 2
go

Another terminal: pgrep -wg $(pidof lc0) | xargs -n1 taskset -cp

[Menkib64]

There is now windows code. I tested win 10 version but I couldn't test win 11 code paths. They are untested.

[borg323]

Returning a disabler object here will simplify the logic in the top level meson.build.

[Menkib64]

Windows 11 returned an unknown error when setting thread affinity. I will have to figure out what caused the problem.