HER: Human-like Reasoning and Reinforcement Learning for LLM Role-playing