[llvm] [AMDGPU][NewInsertWaitcnts] Boilerplate for the new wait insertion pass (PR #226335)

[Pierre van Houtryve via llvm-commits]

https://github.com/Pierre-vh commented:

Also, please add a bit more content to the commit description explaining how we are co-developing it and how development will proceed. I think this is a good overview:

1. First milestone is getting conservatively correct waitcnts (maybe only for a subset of targets). 
2. Over the next few months, iteratively improve the implementation:
    - We refine the architecture and may rewrite/redesign some parts of this until we are  satisfied with the implementation. 
    -  We will also continue progressing towards full parity with the old pass at the same time, addressing remaining targets/edge cases one by one.
3. Permanently switch to the new pass.

https://github.com/llvm/llvm-project/pull/226335