Remove our RDTSC code?

[RalfJung]

In #4524 we gained some code to directly access the x86 RDTSC clock for profiling purposes. However, on almost all systems, std::time::Instant should already be using that clock. So maybe instead of carrying such low-level code we should document how to diagnose and potentially work-around the case where the kernel doesn't trust the RDTSC clock? If the clock is so unstable that the kernel doesn't use it, maybe it's also not a great fit for profiling...

See the discussion here.
Cc @Stypox

[bjorn3]

std::time::Instant goes through libc, which in turn likely goes through the VDSO which finally may use rdtsc. This is more overhead than directly using rdtsc, as libc and the VDSO can't be inlined.

Edit: Never mind, the comment above the linked one says that the overhead is pretty small.