[llvm] Revert "[IndVarsSimplify] sinkUnusedInvariants is skipping instructio… (PR #184436)

[Vigneshwar Jayakumar via llvm-commits]

VigneshwarJ wrote:

> I don't think this change should be reverted. We might remove the transform entirely if it's not, on average, beneficial to performance, but we should not go back to a version of it that will only try to apply sinking to some subset of instructions due to an implementation bug.

Yes, Thats something I want to do.  I have a patch for removing this entirely, #169250

https://github.com/llvm/llvm-project/pull/184436