Threadpool - CPU topology detection

[mratsim]

Followup on #293

• Outline topology detection goals:
  • get_num_threads_logical: with no OS/C stdlib dependency on x86 (all other arch need privileges to access specific registers)
  • get_num_threads_physical: with no OS/C stdlib dependency on x86 (all other arch need privileges to access specific registers)
  • get_num_threads_os: takes into account disabled cores (Virtual machines)
  • get_num_threads_process: Takes into account cgroup, schedaffinity and taskset.

Goals:

• Allow counting physical and/or logical cores without syscalls/IO on Linux x86 (no /proc or /sys read) for hardened trusted enclaves like SGX and full no C stdlib support.
• Allow counting available cores on Linux when running constrained by cgroups, CPUSET, taskset, scheduler affinity, especially in LXC, Docker and Kubernetes containers.

Unfortunately I don't think there is a way to detect more granular time quotas (in Docker/Kubernetes) as opposed to just restricting available CPUs.

See https://github.com/mratsim/constantine/blob/55785cdb65434933e3b7a5f822262201cb46ebf1/constantine/threadpool/primitives/topology.md