[llvm] [ExpandMemCmp][AArch64][PowerPC][RISCV][X86] Use llvm.ucmp instead of (sub (zext (icmp ugt)), (zext (icmp ult))). (PR #121530)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon approved this pull request.

x86 is acceptable, although we might need a peephole to convert any residual movsbl back to sbbl+xor 

https://github.com/llvm/llvm-project/pull/121530