Evolutionary Augmentation Policy Optimization for Self-supervised Learning