Route Spinlock.c's three MSVC intrinsics through the platform
intrinsics layer instead of the Windows builtins:
_mm_pause -> CpuPause (already existed)
_interlockedbittestandset -> CpuInterlockedBitTestAndSet (new)
InterlockedCompareExchange-> CpuInterlockedCompareExchange (new, 32-bit)
Add the two new Cpu* wrappers to the kernel PlatformIntrinsics
{.h,.c} with win + linux arms (Linux uses __atomic_* builtins,
mirroring the existing 64-bit CAS), keeping the Cpu* wrappers the
single intrinsic boundary.