Use SIMD equality for PartialEq on SIMD vectors

[sstangl]

A small change implementing parallel PartialEq for Simd vectors, fixing a TODO. All Simd vectors by definition support SimdPartialEq, so the logic is copied.

I currently think that the call to simd_reduce_all should indeed be round-tripped through Mask to produce a proper integer vector, to handle odd architectures. I wasn't sure if simd_reduce_all(simd_eq(x, y)) was valid, in particular because the definition in LLVM appears to be bitwise AND, and I wasn't clear on where the reduction to Rust bool takes place.

[workingjubilee]

I believe currently simd_eq returns a vector of integers where the values are all -1 or 0. I suspect this is more correct code in any case, at least, relative to possible implementation variances in the future.

[clarfonthey]

It looks like this is kind of the same #274? I personally would prefer the one that uses the safe methods, rather than the intrinsics directly, but they appear to do the same thing.

[workingjubilee]

Do you mean #209? Yes, it effectively is.