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

[Akash Dutta via llvm-commits]

akadutta wrote:

> Eyeballing the test, it doesn't look like we have test coverage for all the permutations of scalar / vector / constant in each operand position. Given the pattern complexity we probably should test all of these

I added a few new tests with immediates in different arg positions. 

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