Text this: Evaluating LLMs for visualization generation and understanding