Will LLMs Replace the Encoder-Only Models in Temporal Relation Classification?