Prompting Large Language Models for Clinical Temporal Relation Extraction