[llvm] [X86] Combine `uitofp <v x i32> to <v x half>` (PR #121809)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon commented:

I'd prefer this was handled in VectorLegalizer::ExpandUINT_TO_FLOAT since the issue isn't x86-specific

https://github.com/llvm/llvm-project/pull/121809