Finer: Investigating and Enhancing Fine-Grained Visual Concept Recognition in Large Vision Language Models