[llvm] [AMDGPU] Skip s_delay_alu for gfx1250 WMMA C-reuse chains (PR #214101)

[via llvm-commits]

https://github.com/sstipano commented:

Changes LGTM overall.

However, I think this should be enabled for GFX12+ and for all WMMA instructions, not just XDL WMMA. Shouldn't be much work, just to change 
`ST->hasGFX1250Insts()` -> `ST->hasGFX12Insts()`
`SII->isXDLWMMA(MI);` -> `SII->isWMMA(MI)`
and couple additional test cases.

cc @rampitec @jayfoad 

https://github.com/llvm/llvm-project/pull/214101