Simulate real-time text streaming from an LLM