Add RDNA3 BUFFER_ATOMIC_ADD_F32 -> interlockedAddf32 support (and all the fp32 atomics)

[Bloodwyn]

Dear AMD folks,

It has been quite some time since the release of the RDNA3 architecture, which has support for atomic floating-point instructions.
So far, they are only exposed to other APIs such as Vulkan, but NOT to DirectX, which is a shame!

NVIDIA has been supporting it for several in years in their extension - so please also add it.

Cheers

[oscarbg]

Plus now, it seems there are new undocumented RDNA4 AGS intrinsics being used for FSR4 presumably FP8 opcodes?:
See:
HansKristian-Work/vkd3d-proton#2398

7624.558:0128:0178:err:vkd3d-proton:vkd3d_dxil_log_callback: dxil-spirv: Unsupported AGS magic instruction 0x29 (immediate 0).