The Gradient Puppeteer: Adversarial Domination in Gradient Leakage Attacks through Model Poisoning