Improvements for MCP Prompts

I've been playing around with Q CLI's support for MCP prompts and have some feedback for improvements that I think would make a big difference in the overall experience:

### UX low-hanging fruit

1. `/prompts list` should include the prompt description so it's easy to understand what each prompt does
2. There should be something like `/prompts details ` that shows all properties of the prompt returned by the MCP server. Specifically argument details.
3. `/prompt get` causes Q to go into a Thinking loop before outputting the prompt. Is this necessary? Shouldn't fetching the prompt be purely deterministic? Edit: I didn't realize that `/prompt get` not only fetches the prompt but also sends it to the model for processing. So `/prompt get myprompt` is the same as running `@myprompt`. So the UX should be modified to output the prompt so it's clearer that this is what's happening.
4. If you try to call a prompt without passing a required argument, you get this really ugly error with a raw JSON output from the MCP server.

### Feature Request

You should be able to pin a MCP server prompt as context, both in agent configuration JSON and via `/context add`For the last one, picturing being able to do something like this:

```
> /context add @some-mcp-prompt arg
```

This will expand the prompt and pin it to context. For agent configuration, you could support something like `prompt:///...` in addition to `file://` supported today. Not quite sure how friendly the URL format will be for prompt arguments, but you could do get params encoded in the URL although depending on the prompt, that could get very long...