Revitalizing CNN Attention via Transformers in Self-Supervised Visual Representation Learning