Learning Symbolic Rules over Abstract Meaning Representations for Textual Reinforcement Learning