[VectorCombine] foldShuffleToIdentity can't handle BITCAST nodes

[RKSimon]

Noticed on #96882 - SSE/AVX code tends to result in a lot of bitcasts as the __m128i / __m256i / __m512i types are always treated as vXi64.

CC @davemgreen

[davemgreen]

Ah interesting. I left bitcasts out as they can make tracking the lanes more difficult (and I honestly didn't think they can up as often as they apparently do from intrinsics).