Upload results/all_perturbation_scores.json with huggingface_hub

Add per-sample E1 truncation sweep results (69,524 records with per-reader correctness at each truncation %)

Upload results/cot_texts.json with huggingface_hub

initial commit