The Potential of Vision-Language Models for Content Moderation of Children's Videos