Manual backgrounding of a subagent

A user should be able to use some keyboard shortcut to background/async a running subagent.

In this use case, we must:

- [ ] Ensure that confirmations from the async subagent are still shown to the user
- [ ] Ensure there is some visual indication for a user that an agent is running in the background
- [ ] When the subagent finishes, the results from the subagent should be naturally pulled into the context of the main agent. This could involve a system reminder appended to function responses

The exact mechanism to return context/feedback should be thought about. It could be a complete return at a correct moment in the ReAct loop, a simple context injection for the main model to go check a output log of the results that it can read, etc.