CAP: Controllable Alignment Prompting for Unlearning in LLMs