Boosting Unsupervised Contrastive Learning Using Diffusion-Based Data Augmentation From Scratch