Minutiae-Guided Fingerprint Embeddings via Vision Transformers