Implement score_answer() for HanoiDataset (Tower of Hanoi)

[andreaskoepf]

In the first version of the Tower of Hanoi task dataset the HanoiDataset class generates a list of instructions as answer. This list cannot be correctly processed by the default implementation of score_answer() in the ProceduralDataset base class. Beside the optimal solution (minimum number of moves) a non-optimal solution (leading to correct end configuration) should also get a non-zero score.

Override the score_answer() method in HannoiDataset and compute a score based on correctness of the end-configuration and number of steps (e.g. compared to optimal solution).

[andreaskoepf]

(@joenorton can probably work on this tomorrow)