African or European Swallow? Benchmarking Large Vision-Language Models for Fine-Grained Object Classification