bmutil.h: Simply go with <x86intrin.h> under (non-Windows) x86_64.

[ucko]

At least under macOS, it was possible to get errors that _mm_popcnt_u{32,64} were undeclared; another branch would have failed due to direct inclusion of , which insists on being included via . is overkill, but still not all that heavy, and avoids any such complications.