Continual Reinforcement Learning by Planning with Online World Models