[llvm] [VectorCombine] Add Cmp and Select for shuffleToIdentity (PR #92794)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

@davemgreen I'll take a look at this - we have a few cases on AVX1 where we don't split 256-bit vectors when we probably should.

https://github.com/llvm/llvm-project/pull/92794