Enhancing Long Chain-of-Thought Reasoning through Multi-Path Plan Aggregation