[llvm] [MISched] Use right boundary when trying latency heuristics (PR #116592)

[Pengcheng Wang via llvm-commits]

wangpc-pp wrote:

> > I suppose this doesn't cause a crash, do you have any example that shows inferior scheduling, which potentially can be a test case?
> 
> do you by any chance have a test?

Not yet. Actually I did this change not because I met some problems. I did find a problem when handling clustering/fusion in post-ra scheduling, which should be fixed by #116584. Then I just checked other heuristics to see if there may be some other issues, and I found this one. I don't know if I can write a test, the problen here and the fix are clear (we should use the right boundary for different directions).

https://github.com/llvm/llvm-project/pull/116592