Stanford Researchers Released AgentFlow: In-the-Flow Reinforcement Learning RL for Modular, Tool-Using AI Agents