Improving LLM-based Ontology Matching with fine-tuning on synthetic data