[llvm] [AMDGPU] Fold operand after shrinking instruction in SIFoldOperands (PR #68426)

[Joe Nash via llvm-commits]

Sisyph wrote:

LGTM, but please wait for rampitec. This is a good approach, though it was nice to be able to remove folding from the shrinking pass https://github.com/llvm/llvm-project/pull/68045/ . Is it still possible with this patch instead of https://github.com/llvm/llvm-project/pull/67878?

https://github.com/llvm/llvm-project/pull/68426