Add a `/pause` or `/stop` slash command to gently interrupt an agent

### Describe the feature or problem you'd like to solve

I want a way to gently interrupt an agent the next time it finishes reasoning. I hate canceling it mid-reasoning, but sometimes I need it to pause so we can talk.

### Proposed solution

There are times when Copilot is doing lots of reasoning and moving really quick. I want it to just pause, so I can get some context of what's going on or take a moment to think through what's going on. I don't need it to cancel and just be stopped and cut off mid-reasoning token. I also don't always need to just give it a nudge. And it feels a little bit weird to send a response of "stop" that the LLM then has to reason about to stop. If we added a `/stop` or `/pause` it could just stop copilot at the next time it tries to take an action. It can complete it's thought so it's in the context of the conversation, but I can then resume control. By me responding, it can go back to work. Not sure if you need any other command to resume (probably not?).

### Example prompts or workflows

_No response_

### Additional context

_No response_

No files, tests, or entry points are identified in the issue, so first locate the CLI's command handling and agent interruption flow. Clarify whether `/pause` or `/stop` is preferred, when interruption occurs, and how work resumes; done should match the agreed command behavior without cutting off the current reasoning.

Written by the indexing model from the issue text.