Coarse-to-Fine Process Reward Modeling for Mathematical Reasoning