Similar Items: Collective Alignment in LLM Multi-Agent Systems: Disentangling Bias from Cooperation via Statistical Physics