"They are uncultured": Unveiling Covert Harms and Social Threats in LLM Generated Conversations