feat: update category matching algorithm

[raphael0202]

What

• See Too many incorrect predictions with Elasticsearch category predictor #918

Part of

• Category prediction (tracker) #379

[codecov]

Codecov Report

Merging #952 (c297f99) into master (3a5f9ab) will increase coverage by 0.17%.
The diff coverage is 64.45%.

@@            Coverage Diff             @@
##           master     #952      +/-   ##
==========================================
+ Coverage   53.06%   53.24%   +0.17%     
==========================================
  Files          93       86       -7     
  Lines        7076     6912     -164     
==========================================
- Hits         3755     3680      -75     
+ Misses       3321     3232      -89     

Impacted Files	Coverage Δ
robotoff/cli/insights.py	0.00% <0.00%> (ø)
robotoff/elasticsearch/export.py	84.84% <ø> (-12.38%)	⬇️
robotoff/off.py	38.33% <0.00%> (+3.02%)	⬆️
...rediction/object_detection/utils/label_map_util.py	22.22% <0.00%> (ø)
...object_detection/utils/string_int_label_map_pb2.py	100.00% <ø> (ø)
robotoff/prediction/ocr/image_flag.py	41.02% <ø> (ø)
robotoff/prediction/ocr/label.py	72.30% <ø> (ø)
robotoff/products.py	42.75% <ø> (ø)
robotoff/utils/cache.py	95.45% <0.00%> (ø)
robotoff/cli/main.py	33.94% <16.66%> (-0.51%)	⬇️
... and 30 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[alexgarel]

Great idea and implementation.

I have some comments, most are style, but some might be important.
However I approve and let you fix what you want to fix.

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
8 Security Hotspots
140 Code Smells

0.0% Coverage
0.3% Duplication