[llvm] [AMDGPU] Added hot-block-rematerialize pass (PR #136631)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm commented:

Only started skimming, but this is a huge amount of code and disproportionately large compared to the new tests. Why is this done while maintaining LiveIntervals in codegen? We know which blocks are hot in the IR, can't we just do trivial block cloning there? 

https://github.com/llvm/llvm-project/pull/136631