[llvm] [AMDGPU] prevent shrinking udiv/urem if either operand is in (SignedMax,UnsignedMax] (PR #116733)

[via llvm-commits]

choikwa wrote:

Going to try to use computeKnownBits for signed and unsigned cases for consistency and wider available information. Will also look at the LibC failure.

https://github.com/llvm/llvm-project/pull/116733