CopyLens: Dynamically Flagging Copyrighted Sub-Dataset Contributions to LLM Outputs