Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation