[llvm] AMDGPU: Handle folding frame indexes into s_add_i32 (PR #101694)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm commented:

It turns out I relaxed the restriction in #101691, and this now does make an observable change. SIFoldOperands can now introduce frame index uses in other instructions 

https://github.com/llvm/llvm-project/pull/101694