Full sim eval rollouts

Every rollout behind the zero-shot chart: 27 unseen robots and YAM, the robot Dream-1 was trained on, 297 rollouts each in simulation. 147 rollouts cover 49 language-following instructions, three runs each; 150 cover 150 unseen tasks, one run each. Every rollout was graded 0, 0.25, 0.5, 0.75 or 1 for how far it got; the progress score is the mean grade, and the success rate counts only the full successes. Head camera, played at 3x.