[llvm] [AMDGPU] Add v2i32 V_BFI combines for andn2 and orn2 (PR #196325)

[Shilei Tian via llvm-commits]

https://github.com/shiltian commented:

LGTM but I wonder if we can simply just do a bit cast and then use the existing i64 pattern?

https://github.com/llvm/llvm-project/pull/196325