Captures for retraining

Visual Detector saves live frames with their detections as captures.zip. Import it in Model Bench with Import ctrlX Captures.

1. Collect captures on the device.
2. Download captures.zip from the web interface.
3. In Model Bench, choose Import ctrlX Captures. Frames arrive as unreviewed annotations, filed under ctrlx/<device>/cam<N>.
4. Correct, mark reviewed, retrain and export the new model.

Importing the same zip twice adds nothing.