What is the agent-environment loop in reinforcement learning, and how does it describe the interaction between an agent and its environment? How do states, actions, rewards, and feedback work together to help an agent learn better decisions over time?