Computational segmentation of Wayang Kulit video recordings using a Cross-Attention Temporal Model