Cross-Encoder for Unsupervised Gaze Representation Learning