Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning