[llvm] [AMDGPU] Disable atomic optimization of fadd/fsub with result (PR #96479)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> but unfortunately the IR atomicrmw instruction does not have fast math flags.

If we really have to consider this, it's better to rely on the function attributes as a placeholder until we have a per-instruction solution to atomicrmw 

https://github.com/llvm/llvm-project/pull/96479