Shorten eval source label for widget display

Metadata audit: 34-language list, result sources, modelopt/safetensors tags, ShadowRock logo header, intended-use section; drop datasets field

Metadata: inference:false, quantized_by, calibration datasets, metrics

model-index: add NDCG@10 display name to metrics

Expand model-index to all measured tasks

Initial community release: weights, card, scripts, manifests, raw eval results