Text this: Efficient and Stable Learning for Distribution Network Operation: A Model-Based Reinforcement Learning Approach