Lifelong Reinforcement Learning with Modulating Masks