[llvm] [AMDGPU][wmma] - Add tied wmma intrinsic (PR #69903)

[Jessica Del via llvm-commits]

OutOfCache wrote:

> The "Simplify wmma instruction defs" commit is clearly good. Maybe commit that one separately first?

Did that in [this PR](https://github.com/llvm/llvm-project/pull/70622). Closing this PR for now. I will open another one for the new intrinsics after that one is merged.

> Need to add clang builtins too, similarly to [D128952](https://reviews.llvm.org/D128952) (can be a separate patch).

I added them locally and will open a PR once these intrinsics are merged.

https://github.com/llvm/llvm-project/pull/69903