Text this: Hierarchical Support Vector State Partitioning for Distilling Black Box Reinforcement Learning Policies