Show the router's recommended model for "Auto" before executing, with the option to override

When I select "Auto" for model selection, the request runs right away against whatever model the router picks. That works, but I never get to see which model was chosen or step in before it runs.

It would help to have the router return its recommended model for a given query first, then let me either accept it or pick a different model. So "Auto" becomes a recommendation I can confirm or overrule, rather than a decision that happens silently.

## What I'd like

- When "Auto" is selected, the router surfaces its recommended model for the current query before the request executes.
- I can accept the recommendation and run with it, or switch to another model and run with that instead.
- Optionally, a setting to keep the current behavior (auto-run immediately) for people who prefer it.

## Why

The router usually gets it right, but not always. Some queries are cheap enough that I want a smaller model, and some I know need a stronger one regardless of what the router thinks. Seeing the pick first gives me that control without giving up the convenience of Auto.

Start by locating the Auto model-selection flow and the router request entry point. Trace where the recommended model becomes available before execution, then define the accept, override, and optional auto-run behavior; done means Auto no longer runs silently when confirmation is enabled.

Written by the indexing model from the issue text.