[llvm] [AMDGPU] Fix SIFoldOperands miscompiling values that leave a divergent loop (PR #203256)

[Arseniy Obolenskiy via llvm-commits]

aobolensk wrote:

bare SGPR->VGPR COPY with no `implicit $exec` reachable from plain IR through (in both gisel and sdag cases)

I could not find an IR for normal e2e scenario

https://github.com/llvm/llvm-project/pull/203256