Method-Based Reasoning for Large Language Models: Extraction, Reuse, and Continuous Improvement