Value-Distributional Model-Based Reinforcement Learning