Protecting Users From Themselves: Safeguarding Contextual Privacy in Interactions with Conversational Agents