The Lessons of Developing Process Reward Models in Mathematical Reasoning