Use mm_cmpgt_epuXX for unsigned integer comparison on AVX512

[wolfv]

AVX 512 added some SSE and AVX2 sized instructions for comparing unsigned integers.

The catch is that they return a __maskX and not a SIMD register size batch of bool ... so far we don't handle that.

But we could remove the xor flipping code that we currently need in order to use the signed comparisons for unsigned integers.