VRPRM: Process Reward Modeling via Visual Reasoning