[llvm] [AMDGPU] Add option to prevent insns straddling half cache-line boundaries (PR #150239)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> The strategy seems wrong, this should be handled as part of relaxation

i.e. the solution should not involve calling getInstSizeInBytes. That can never be 100% accurate, and you need to account for other relaxations you also cannot know without also being part of the relaxation process 

https://github.com/llvm/llvm-project/pull/150239