Teaching LLMs Human-Like Editing of Inappropriate Argumentation via Reinforcement Learning