- Introduce a canvas-based audio visualizer overlay on opponent portraits that activates during speech playback
- Update `SpeechQueue` to support `onStart` and `onEnd` callbacks for hooking into audio events
- Modify `Portrait` to trigger the visualizer with emotion-specific colors (blue for intro, green for happy, red for upset)
- Add `speech` metadata to `opponents.json` for Ethel and Bernie, defining available intro/happy/upset audio clips
- Ensure proper cleanup of visualizer state and resources on portrait destruction
