$\textit{LinkPrompt}$: Natural and Universal Adversarial Attacks on Prompt-based Language Models