Generalizing Consistency Policy to Visual RL with Prioritized Proximal Experience Regularization