Synthetic Knowledge Ingestion: Towards Knowledge Refinement and Injection for Enhancing Large Language Models