[llvm] [AMDGPU] Add pattern to select scalar ops for fshr with uniform operands (PR #165295)

[Akash Dutta via llvm-commits]

akadutta wrote:

@jayfoad I've enabled it for all archs and addressed the other comments. It does seem to impact GFX6,7, and 8. Please take a look if this seems okay.

https://github.com/llvm/llvm-project/pull/165295