Exposing Pink Slime Journalism: Linguistic Signatures and Robust Detection Against LLM-Generated Threats