Bootstrapping LLM-based Task-Oriented Dialogue Agents via Self-Talk