[llvm] [AMDGPU] Stop optimising readfirstlane in pass AMDGPUUniformIntrinsicCombine (PR #166955)

[Tracy Sharpe via llvm-commits]

raiseirql wrote:

> @raiseirql please could you test this?

@jayfoad @PankajDwivedi-25 I merged this change to our branch and yes, the codegen looks good and our use case runs without crashes/errors.

https://github.com/llvm/llvm-project/pull/166955