ChatGPT's Dangerous Sycophancy: How AI-Bootlicking Reinforces Mental Illness