[llvm] [indvars] Prove non-negative for widening IVs in count down loops (PR #71214)

[Philip Reames via llvm-commits]

preames wrote:

I found another approach to these same cases over in https://github.com/llvm/llvm-project/pull/71557.  My current thinking is that both are valuable, but that change is probably more obvious.  I suggest reviewing that one first, and then I'll need to figure out some more extensive test coverage to justify the complexity of this one.

https://github.com/llvm/llvm-project/pull/71214