[AndersenAgnostic] Backport fixes and expanded statistics for Andersen analysis

[haved]

During work on the andersen-no-flags branch, some other changes were made as I discovered issues or wanted more statistics:

• There was a bug where Hybrid Cycle Detection could invalidate an iterator while using it.
• The topological Worklist algorithm has been made slightly less terrible by actually tracking work items now, and only visiting nodes that need to be visited during the sweep. The ObserverWorklist is now called the WorkSet instead, as that name describes its new job (an unordered worklist) better. This is likely closer to what has been done by Pearce07, but it still does not make it the best worklist iteration order.
• Make the JLM_ANDERSEN_TEST_ALL_CONFIGS environment variable take an integer controlling the amount of iterations each config should be used.

The bulk of this PR is related to getting more statistics:

• Count attempts at growing points-to sets, as well as how many items are removed from points-to sets during solving (e.g due to unification or PIP).
• Add code to the PointsToGraph class to count the number of edges, as well as the number of pointer-pointee relations, since RegisterNodes can represent multiple registers at once.
• Add the flag CanPoint() to PointerObjects to formalize the existence of memory objects whose points-to sets we do not track (such as int a). This makes the statistics more useful and comparable, as well as removing potentially tons of edges in the PointsToGraph going out from these memory objects.
• Add statistics for counting pointer-pointee relations to be able to say something about precision, regardless of unification.

This PR ended up being a bit larger than I wanted, let me know if you want it split up