Rethinking Robust Contrastive Learning from the Adversarial Perspective