NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning