MVL-SIB: A Massively Multilingual Vision-Language Benchmark for Cross-Modal Topical Matching