Reinforcement Learning for Large Language Models: A Deep Dive