Partial Models for Building Adaptive Model-Based Reinforcement Learning Agents