[llvm] [MachineSink] Sink into consistent blocks for optsize funcs (PR #115367)

[Steven Johnson via llvm-commits]

steven-johnson wrote:

(I'm not sure how to give you a repro for this -- I've only seen it happen once, on a Google-proprietary JIT test that I don't think I can easily provide IR for. That said, assuming the diagnostic re: strict-weak-ordering is correct, it should be simple to verify and fix via inspection...)

https://github.com/llvm/llvm-project/pull/115367