Selecting and Combining Large Language Models for Scalable Code Clone Detection