headers: Implement the _bittest* intrinsics for arm and aarch64

These are non-atomic versions corresponding to the existing
_interlockedbittest* - therefore they're implemented as plain C
instead of assembly. Additionally, there are plain intrinsics "_bittest"
and "_bittest64", where there's no corresponding interlocked version.

Signed-off-by: Martin Storsjö <martin@martin.st>