Feature: Apply sparse-checkout at session worktree creation to avoid full-checkout timeouts on large repos

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

Apply sparse-checkout at session worktree creation to avoid full-checkout timeouts on large repos

### Proposed solution

## Summary

When the app creates a new project session, it runs `git worktree add`, which
checks out the **entire** working tree. For very large monorepos this checkout
is slow enough to **time out**, so sessions fail to start.

New worktrees do **not** inherit sparse-checkout config from the main repo, and
hooks (e.g. `PreToolUse`) only fire *after* the worktree exists — so there is
currently no way to avoid the full checkout from the user side.

## Problem

- Repo: very large local repo (working tree in the multi-GB / high-file-count range)
- `git worktree add ` materializes the full tree
- Worktree creation times out → session cannot be created
- No pre-worktree hook / config exists to scope the checkout

## Proposal

Add a config option to scope the checkout when creating session worktrees, e.g.:

```jsonc
// settings.json
{
"session": {
"worktree": {
"sparseCheckout": {
"enabled": true,
"cone": true,
"paths": ["src/app", "libs/shared"]
}
}
}
}
```

### Example prompts or workflows

_No response_

### Additional context

Prior art / related
#2399 — sparse-checkout for plugin installs
#2463 — sparse-checkout for marketplace clones
Both apply sparse-checkout to reduce clone cost, but only for plugins/marketplace,
not for session worktrees. This request covers the session-creation path.

Environment
Copilot CLI version: 1.0.71-2
OS: Windows
Repo: local-only, very large working tree

Beginne beim Pfad zur Erstellung des Session-Worktrees und beim Aufruf von `git worktree add`, und ermittle dann, wie die Konfiguration vor dem Checkout angewendet werden kann. Die Aufgabe ist erledigt, wenn ein Session-Worktree die vorgeschlagenen sparse-checkout-Einstellungen verwenden kann, sodass große Repositorys den vollständigen Baum nicht materialisieren und keine Zeitüberschreitung verursachen.

Vom Indexierungsmodell aus dem Issue-Text verfasst.