MSR: Making Self-supervised learning Robust to Aggressive Augmentations