DisCoRL: Continual Reinforcement Learning via Policy Distillation