Improving Robustness of Vision Transformers via Data-Augmented Virtual Adversarial Training