[llvm] [AMDGPU] Fix wrong reverse operations for `v_cmp_*` (PR #146398)

[Shilei Tian via llvm-commits]

shiltian wrote:

This PR is still WIP, as I still have several test cases to update manually. However, since the code I'm modifying here has existed for nearly a decade, I'm a bit nervous about potentially missing some historical context. From my current understanding, it should be `lt -> ge` and `le -> gt`.

https://github.com/llvm/llvm-project/pull/146398