NaVILA: Legged Robot Vision-Language-Action Model for Navigation