SQ-LLaVA: Self-Questioning for Large Vision-Language Assistant