[llvm] [VectorCombine] Fold (icmp eq/ne (reduce.add X), 0) to reduce.umax (PR #180001)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon requested changes to this pull request.

Ensure the new tests are committed to the patch stack first so the later commits can show the codegen difference

https://github.com/llvm/llvm-project/pull/180001