RAG-4: Automated doc refresh via scheduled CI

[nonogv]

Context

Depends on #66 (RAG-1). Part of the v2 RAG milestone.

What

Add a scheduled GitHub Actions workflow that keeps data/live-docs.json current without manual intervention.

File: .github/workflows/refresh-live-docs.yml

```yaml
on:
schedule:
- cron: '0 4 * * 1' # every Monday 04:00 UTC
workflow_dispatch: # manual trigger from Actions tab
```

Workflow steps:

1. Checkout repo
2. npm ci
3. npm run scrape-docs
4. If data/live-docs.json or data/live-docs-diff.json changed:
   • Open a PR: title chore: refresh Live docs index (YYYY-MM-DD)
   • PR body: populated from data/live-docs-diff.json (added/removed/updated/unchanged)
   • Auto-assign repo owner as reviewer
5. If nothing changed: exit cleanly — no PR, no noise
6. If scraper exits non-zero: mark workflow as failed (GitHub sends default failure notification)

Stale index fallback (for fork builds):

src/agent/docs-index.ts already exports isDocIndexStale() (from RAG-2). The UI shows a dismissible diagnostic message when it returns true and data/live-docs.json exists. This only affects developers running from a stale fork; packaged .vsix builds always ship with the index current at release time.

Resilience note

The scraper should keep old chunks rather than deleting them when a page returns 404 — log a warning instead. A single ableton.com restructure should not silently wipe the index on a Monday.

Acceptance criteria

• Workflow runs on schedule and on workflow_dispatch
• A PR is opened only when content has changed
• PR body shows the diff summary from data/live-docs-diff.json
• A scraper failure marks the workflow run as failed
• Old chunks are preserved (not deleted) when their source page returns 404