Mitigating Bias in Queer Representation within Large Language Models: A Collaborative Agent Approach