FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation | Read Paper on Bytez