AquaVLM: Improving Underwater Situation Awareness with Mobile Vision Language Models