feat: add yolo model for nutriscore object detection

[raphael0202]

Integrate first Yolov8 object detection model, for Nutri-score detection: https://huggingface.co/openfoodfacts/nutriscore-yolo.

This model runs in ~50ms, which is several order of magnitude faster than the current Faster-RCNN ResNet-101 model that we have in production.
If successfully deployed, the idea is to retrain all TF Object Detection API models with Yolov8.

[codecov]

Codecov Report

Attention: Patch coverage is 9.43396% with 96 lines in your changes missing coverage. Please review.

Project coverage is 56.41%. Comparing base (2d00822) to head (d810ba5).

Files	Patch %	Lines
robotoff/prediction/object_detection/core.py	7.21%	90 Missing ⚠️
...tion/object_detection/utils/visualization_utils.py	0.00%	4 Missing ⚠️
robotoff/prediction/upc_image.py	0.00%	1 Missing ⚠️
robotoff/utils/image.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1356      +/-   ##
==========================================
- Coverage   57.32%   56.41%   -0.92%     
==========================================
  Files          79       79              
  Lines        7850     7927      +77     
==========================================
- Hits         4500     4472      -28     
- Misses       3350     3455     +105     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.