Memory leak when running a high number of solver phases

Tech: AcceptedForager doesn't clean up after itself in the phaseEnded() method.
As a result, the entire working solution of a phase can be kept in memory. When having a lot of phases (which is very unusual unless your result to partititioning) this hurts. In combination with a very big dataset (such as my xmas holiday hobby packing-santas-sleigh competition), this causes OutOfMemoryError.

Most (if not all) users use 2 or 3 solver phases, so the death of the Solver already releases that memory before it ever turns into a problem.

Testing the fix now on master. I suggest we backport this to 6.0.x as the fix is trivial and harmless.