MSTS: A Multimodal Safety Test Suite for Vision-Language Models