Token Boosting for Robust Self-Supervised Visual Transformer Pre-training