[llvm] [AMDGPU] Factor out common code from SIInsertWaitcnts (PR #83018)

[Christudasan Devadasan via llvm-commits]

cdevadas wrote:

> though it's hard to judge the usefulness of this without an in-tree use case.
I see your point. This refactoring moves the code into new files and this is immediately required in a downstream branch.
If this patch isn't upstreamed now, any subsequent waitcnt work in the upstream will be hard during the downstream merge and sometimes won't be scalable. 



https://github.com/llvm/llvm-project/pull/83018