Text this: Multimodal skin disease classification using vision transformers, medical captioning, and metadata fusion