Babel-ImageNet: Massively Multilingual Evaluation of Vision-and-Language Representations