LOVM: Language-Only Vision Model Selection