ShareLock Ultra-Lightweight CLIP-like Vision-Language Model SteerViT Steering Vision Representations with Language