Scaling of Search and Learning: A Roadmap to Reproduce o1 from Reinforcement Learning Perspective