IFViT: Interpretable Fixed-Length Representation for Fingerprint Matching via Vision Transformer