[llvm] [InstCombine] Fold `trunc nuw/nsw (x xor y) to i1` to `x != y` (PR #90408)

[via llvm-commits]

asmok-g wrote:

Heads-up: we're seeing a probable mis-compile after this commit in a google internal test. The function that changes behavior with and without optimization is: https://github.com/google/arolla/blob/65c8389f271b9dce973801074217afb0e35f850c/arolla/memory/simple_buffer.h#L336.

Still checking if this is some UB in the code or a real mis-compile.

https://github.com/llvm/llvm-project/pull/90408