[llvm] [AMDGPU] Fix missing S_WAIT_XCNT with multiple pending VMEMs (PR #166779)

[Jay Foad via llvm-commits]

jayfoad wrote:

Eager ping - I tried hard to make this easy to review. The topmost commit should be an obvious bug fix, only clear pending VMEM events if LoadCnt == 0. The other commits are all NFC refactorings leading up to the fix.

https://github.com/llvm/llvm-project/pull/166779