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

[Alexander Kornienko via llvm-commits]

alexfh wrote:

> > Could UBSan flag pointer arithmetic overflows?
> 
> UBSan should already flag this.

We have found a couple of dozen of issues uncovered by this optimization, but I don't think in any of these cases ubsan was producing an error.

So far the only reasonable way to investigate these problems was to selectively disable optimizations using `-O0` for subsets of files and then to bisect within the file by inserting `#pragma clang optimize off`. After that only manual inspection.

Not having a good way to localize these problems is a serious concern for large codebases.

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