Continual Reinforcement Learning with Diversity Exploration and Adversarial Self-Correction