Narrowing by fixed number of meets

[sim642]

Based on the delayed widening from #1483, we could easily implement a domain lifter for narrowing that applies meet at most a fixed number of times. This makes it possible to have a terminating narrow for domains that do not have narrowings. It is a standard technique.

Our current analysis for the polyhedral example that #1484 is fixing is even more imprecise than the old literature suggests. This is because Apron doesn't provide narrowing for polyhedra and thus we cannot make any decreasing iterations. But the standard implementation of narrowing for polyhedra is to do a fixed number of meets.

• Maybe this benefits ApronAnalysis doesnt work on certain loops #1102?