weights: pure-PyTorch port of Sense-X/uniformer_video s K400 16x8 (from f9448914 via scripts/convert_uniformer_video.py)

data: acaua_task=video_classification for task routing

data: Kinetics-400 labels (400 classes, upstream MIT)

docs: add Apache-2.0 LICENSE per s4(a)

docs: add NOTICE attribution per Apache-2.0 s4(d) (code AND weights)

docs: acaua mirror model card with code+weights provenance

initial commit