Still Not Quite There! Evaluating Large Language Models for Comorbid Mental Health Diagnosis