[clang] [analyzer] Prune infeasible states related to concrete ints early to fix a crash (PR #210912)

[Arseniy Zaostrovnykh via cfe-commits]

necto wrote:

> Right, I should have expressed what I mean by updating the title. IMO the main product here is that we touch the solver - that it fixes a crash that materialized in a checker is just a second degree side-effect. The fact that the checker happens to be a pthreadlock, is a 3rd degree side-effect.
> 
> So I'd probably stress on pruning infeasible states in some way but I don't have more constructive examples. Of course in the description it's a fair game to relate this to the checker.

Reformulated. I think fixing a crash is an important effect from the user PoV. If not for the crash, having these infeasible states linger for a little longer isn't really noticeable.

https://github.com/llvm/llvm-project/pull/210912