InsightSee: Advancing Multi-agent Vision-Language Models for Enhanced Visual Understanding