[llvm] unpack packed instructions overlapped by MFMAs post-RA scheduling (PR #157968)

[Akash Dutta via llvm-commits]

akadutta wrote:

@jrbyrnes @hidekisaito @bcahoon @ronlieb Requesting review on this patch. This is a cleaned up version of PR #151704. The optimization is now applied post RA scheduling in si-pre-emit-peephole instead amdgpu-pre-ra-optimizations.

https://github.com/llvm/llvm-project/pull/157968