SQUAB: Evaluating LLM's robustness to ambiguous and unanswerable questions in semantic parsing