Practical self-supervised continual learning with continual fine-tuning