[llvm] [SCEV] Handle backedge-count logic for std::reverse like loops (PR #92560)

[via llvm-commits]

mrdaybird wrote:

> I was experimenting with the Alive2 proof a bit, and ran into the following which seems a bit worrying: https://alive2.llvm.org/ce/z/gabymq . (See [AliveToolkit/alive2#727](https://github.com/AliveToolkit/alive2/issues/727) for discussion of the unrolling options.)

Oh! This indeed is a bug that needs to be handled. 
I am thinking that, adding a checking for any no-wrap flags for RHSAddRec should suffice at minimum, though there could be more cases that we can handle. What do you think? @efriedma-quic 
Also alive2 also seems to not have any problems if the RHS recurrence has nsw/nuw flags.

https://github.com/llvm/llvm-project/pull/92560