Update README: add eval instructions with subset LMDBs and results table

Fix pred_timestamp: store as clip-relative (no double subtraction of clip_s)

Remove narr/nlq base records from train: keep only feedback splits

Update dataset card: document query and feedback fields

Add query/feedback fields: query=NLQ question, feedback=user feedback text

Remove goalstep_q and hd_epic_q configs from dataset card

Remove goalstep_q and hd_epic_q (base NLQ splits not part of EM-QnF)

Remove old .json files, keep only .jsonl

Upload folder using huggingface_hub

Upload README.md with huggingface_hub

Upload folder using huggingface_hub

Upload README.md with huggingface_hub

Upload folder using huggingface_hub

initial commit