[llvm] [DeadStoreElimination] Visit blocks in RPO traversal when initializing DSEState (PR #137815)

[Antonio Frighetto via llvm-commits]

antoniofrighetto wrote:

>From the test suite results, it seems reversing the order leads overall to more improvements than regressions, though I'm not sure if we happen to eliminate more stores incidentally by luck (i.e., reaching the cut-off later); thus, whether this is worthwhile (provided that the change is semantically correct).

https://github.com/llvm/llvm-project/pull/137815