Evaluating Interpretable Reinforcement Learning by Distilling Policies into Programs