SLearnLLM: A Self-Learning Framework for Efficient Domain-Specific Adaptation of Large Language Models