Mitigating Adversarial Perturbations for Deep Reinforcement Learning via Vector Quantization