EXPLORER: Exploration-guided Reasoning for Textual Reinforcement Learning