DriveWorld-VLA: Unified Latent-Space World Modeling with Vision-Language-Action for Autonomous Driving