previews [pushT]: force disc moved onto the camera views at the sensor's projected position (was pinned to the GelSight tile) and shrunk to annotate rather than cover; forces recomputed with react_calib — the previous newton values are void (end-to-end rho 0.143 from pixel-unit weights on mm-unit features)
README: document the estimated contact-force columns (force_*_normal_n / penetration_mm / target_pose), the 1 N/mm stiffness assumption, and the limits — the data shipped without a word about it
