From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection