Support dynamic `sched::CpuSet` size for machines with >64 cores

[gsnoff]

Currently the CpuSet struct consists of a single cpu_set_t (or cpuset_t on *BSD), and sched_*etaffinity syscalls are called with constant cpusetsize parameter, always being mem::size_of::<CpuSet>() (=64 on all targets, at least those I care about) regardless of what the value of sysconf(_SC_NPROCESSORS_CONF) would be.

What I would like to see is some future-proof design, that would support for a runtime-sized allocation of cpu_set_t slice, similarly to what C macros such as CPU_ALLOC() facilitate.

For example, I can see this being abstracted behind CpuSet being an enum internally, that either has some default sized stack value, or dynamically sized slice on heap, and the affinity calls picking the logic depending on the variant. Or there could be two types like CpuSet and CpuSetDyn implementing the same (unsafe?) trait, providing cpusetsize and a pointer to the first cpuset_t, making the affinity fns generic over this trait.

[SteveLauC]

Yeah, this is something we should implement, limiting the cores to 64 makes no sense to me (especially when this has already been supported by the underlying syscall).

I can see this being abstracted behind CpuSet being an enum internally, that either has some default sized stack value, or dynamically sized slice on heap,

This approach looks good to me

[SteveLauC]

Filed PR #2590 for this