[llvm] AMDGPU: Remove flat/global fmin/fmax intrinsics (PR #105642)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I need to be able to call this from Clang code gen. The reason I insist on this is because I don't think the selection of attributes to pass to the current atomicRMW function is clear enough i.e. if I was relying on the intrinsic up until now, it is not clear what set of additional options to pass to atomicRMW() to make it behave like the intrinsic. 

This is exactly why we must remove this intrinsic or anything that acts like it. You, the frontend writer, must semantically know what you can do. The semantics cannot be "do whatever the intrinsic happens to do on this target"

https://github.com/llvm/llvm-project/pull/105642