While working on issue1208, we removed the function convert_ancestor_state, which was previously called in all heuristics. As a side effect, this function used to unpack the state passed to a heuristic. This unpacking was critical for some heuristics - the diverse potential heuristics were most notable in the experiment. We kept the old behavior by unpacking all states in the heuristic base class.

However, some heuristics might actually benefit from working on packed data. In this issue, we want to figure out if a per-heuristic setting for unpacking is worth it and if so how it should be set for each heuristic.