Self-Correction Bench: Uncovering and Addressing the Self-Correction Blind Spot in Large Language Models