[clang] [llvm] [AMDGPU] Track tensor load/store DMAs with asyncmark (PR #200775)

[Ryan Mitchell via cfe-commits]

RyanRio wrote:

Petition to remove 

```
def __builtin_amdgcn_s_wait_asynccnt : AMDGPUBuiltin<"void(_Constant unsigned short)", [], "gfx1250-insts">;
def __builtin_amdgcn_s_wait_tensorcnt : AMDGPUBuiltin<"void(_Constant unsigned short)", [], "gfx1250-insts">;
```

now? Any opposed? @krzysz00 @arsenm @ssahasra @shiltian Any folks not using asyncmark already really should be.

https://github.com/llvm/llvm-project/pull/200775