[llvm] [SCEV] Prove max(X0..Xn) Pred RHS via per-operand decomposition. (PR #208546)

[Florian Hahn via llvm-commits]

fhahn wrote:

> > Does this render #207729 redundant? I believe that one was working around the same SimplifyICmpOperands issue.
> > If we do this, can we drop the non-recursive min/max reasoning, or is it needed as well?
> 
> This is a generalization of the reasoning; this particular patch does not yet make #207729 redundant, but a follow-up will: [19c00de](https://github.com/llvm/llvm-project/commit/19c00de9a4fe4b4d3fc244d685c2686d7df662f5) (part of #208820)

@nikic are you happy with the direction of the set of patches?

https://github.com/llvm/llvm-project/pull/208546