Neighborhoods: Support for non-exhausting iteration

An exhausting iteration is very expensive. For every bi join, we need to keep track of the not-yet-exhausted iterators on both sides; therefore for even mid-sized joins, the memory overhead of that grows beyond what's reasonable.

We want to have iterators which can possibly repeat the value they return. We do not want for the user to have to enable this feature. We think it may have to be enabled automatically in the following situations:

• When we see that the join sizes are getting out of hand.
• When the LS performs a lot of moves per step, because that means the joins would otherwise be building more and more of these expensive iterators.
• For tri-picks and higher, as there we can automatically assume that the joins would be massive.

The issue names no files, tests, or entry points. Start by locating the neighborhood join and iterator implementation, then trace how exhaustion and join size are handled. Done means non-exhausting iterators are supported and the issue's listed situations can enable them automatically without requiring user configuration.