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

[Matt Arsenault via llvm-commits]

arsenm wrote:

> But you're not removing everything that acts like it except for the intrinsic itself. There's nothing illegal about asking for a new style of atomic RMW with a particular set of additional options that happens to match the previous intrinsic.

Part of the point of this is to make you do this work. Adding a convenience to "match the intrinsic" is counterproductive. The intrinsic was not semantically defined 

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