Towards Human Cognition: Visual Context Guides Syntactic Priming in Fusion-Encoded Models