Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection