[llvm] [InstCombine] Support gep nuw in icmp folds (PR #118472)

[Alexander Kornienko via llvm-commits]

alexfh wrote:

My bad, indeed, our cases are associated with invalid pointers rather than specifically with pointer overflow. But I suppose, a similar distribution is to be expected in other code bases, which means that there is no targeted sanitizer for this and diagnostic of issues resulting from this and related changes is non-trivial. Does performance gain justify this?

https://github.com/llvm/llvm-project/pull/118472