Continual Reinforcement Learning with Complex Synapses