Beyond Turn-taking: Introducing Text-based Overlap into Human-LLM Interactions