MultiAgent Collaboration Attack: Investigating Adversarial Attacks in Large Language Model Collaborations via Debate