Official release of PRIMO R1, a 7B video MLLM for robotic process reasoning featuring RL-optimized models, SFT/RL datasets, and cross-domain benchmark