Aligning Large Language Models for Enhancing Psychiatric Interviews through Symptom Delineation and Summarization