[FEATURE] Add token usage warnings for potentially inefficient solution paths (aka: falling down rabbit-holes)

### Preflight Checklist

- [x] I have searched [existing requests](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20label%3Aenhancement) and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)

### Problem Statement

It is quite common that a seemingly simple problem is rather complex.
It is also common that a seemingly complex problem is actually simple.
Claude can sometimes find itself down a massive rabbit-hole, guzzling 10s or even 100s of thousands of tokens to sift through what is essentially junk. This is obviously wasteful and slow.

### Proposed Solution

In cases where a task seemed simple, for example because the user stated it is or after self-reflection by Claude, but a large amount of tokens or many commands are executed Claude should ask the user if they think they are on the right path to a solution or if Claude should stop there and try a different direction.

### Alternative Solutions

Currently I babysit the tasks and `esc-esc` if I find Claude going on odd adventures. This is fine, but it means that background agents or multiple agents is just off the table until I manage to refine the prompts enough to prevent this kind of behavior.

### Priority

Medium - Would be very helpful

### Feature Category

Performance and speed

### Use Case Example

_No response_

### Additional Context

_No response_

No files, tests, or entry points are named in the issue. First locate the command-execution and token-usage paths, then clarify the thresholds, warning interaction, and behavior for background or multiple agents; done should mean that potentially inefficient paths reliably prompt the user to continue or try another direction.

Rédigé par le modèle d'indexation à partir du texte de l'issue.