[llvm] [InstCombine] Add transforms for `icmp uPred (trunc x),(truncOrZext(y))` -> `icmp uPred x,y` (PR #71309)

[via llvm-commits]

goldsteinn wrote:

> Can you please check whether this has compile-time impact (due to computeKnownBits)?
> 
> (We could reuse the KnownBits from foldICmpUsingKnownBits if necessary.)

Seems like minimal affect: https://llvm-compile-time-tracker.com/compare.php?from=2b71f91b06ad4f5a0c54725b06283fd731620b92&to=2343dfccca46d158de240fca8e0588ae4cc46203&stat=instructions:u

Slight regression in ThinLTO. Larger improvement in stage2 + O3.

https://github.com/llvm/llvm-project/pull/71309