[llvm] [AMDGPU] Employ MemorySSA when rewriting out arguments (NFCI) (PR #207922)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> Left a TODO in the code as I’m also unsure whether the pass is (meant to be) semantics-preserving

Yes, this should be a purely mechanical optimization 

https://github.com/llvm/llvm-project/pull/207922