Render bare train queries in the benchmark's test shape; columns id, query, response, scoring_context first

Give train queries their benchmark's reply-format block

Add ground_truth, response and scoring_context beside every existing column; add the all config

Add upstream file provenance; harden the equivalence key

Upload folder using huggingface_hub