π0 and π0-FAST: Vision-Language-Action Models for General Robot Control