[llvm] [AMDGPU][SIInsertWaitcnts] Address post-commit comments for #180864 (PR #183232)

[Sameer Sahasrabuddhe via llvm-commits]

ssahasra wrote:

> > Based on my understanding of this pass, visiting the wait events in a different order really should not cause any correctness problems. I am less sure about whether it could have any effect at all on the generated code.
> 
> @arsenm, do you know if the order makes a difference? I pointed out the fact that the order may be different on the new implementation, but if the order doesn't change anything, I think we should stick with the bitmask.

I don't see why the order should matter. Clearly, `updateByEvent()` is called independently for each event? If order mattered, then there is likely to be a more fundamental issue with how we represent events in the first place.

https://github.com/llvm/llvm-project/pull/183232