[llvm] [LoopUtils] prevent negative estimated trip count result (PR #195610)

[Aiden Grossman via llvm-commits]

boomanaiden154 wrote:

> I have gone off from the IR reproducer, which could technically come from a fuzzer--and in this case I believe the PGO data in the linked issue is deliberately using the negative value. Rejecting negative counts also works.

It looks like the reproducer in the issue came from somewhere later in the optimization pipeline, not directly from the frontend. I don't think `llvm.loop.estimated_trip_count` is emitted outside of some of the loop optimization passes.

https://github.com/llvm/llvm-project/pull/195610