Final viewer ID for accurate unique counts

The processing pipeline now resolves a canonical final_viewer_id per view, and the Data APIs use it everywhere unique-viewer logic runs.

1. The stream processing pipeline picks the last-known viewer ID for a session and propagates it as final_viewer_id on every emitted row.
2. All breakdown, overall and overview queries now count distinct final_viewer_id instead of the raw rotating viewer ID.
3. Produces more accurate unique-viewer counts in sessions where the player rotates its viewer ID mid-session.