HumanCLAW: Can Vision-Language Models Act Through a Body? https://human-claw.github.io/