LaCViT: A Label-aware Contrastive Fine-tuning Framework for Vision Transformers