[llvm] [SCEV] Fix incorrect nsw inference for multiply of addrec (PR #66500)

[Nikita Popov via llvm-commits]

================

----------------
nikic wrote:

I couldn't figure out how to adjust this test and ended up deleting it because it appears to be entirely broken anyway: It was supposed to test that IR not in loop simplify form is handled gracefully, but then ... it explicitly runs loop-simplify, so this is just pointless.

https://github.com/llvm/llvm-project/pull/66500