Empowering NLG: Offline Reinforcement Learning for Informal Summarization in Online Domains