Device-Directed Speech Detection for Follow-up Conversations Using Large Language Models