System Fingerprint Recognition for Deepfake Audio: An Initial Dataset and Investigation