Reject Threshold Adaptation for Open-Set Model Attribution of Deepfake Audio