Scene and animal attributes retrieval from camera trap data with domain-adapted language-vision models