Model-based Lifelong Reinforcement Learning with Bayesian Exploration