feat: add classification module for crop orientation

[charlesmindee]

This PR adds a classification module with a orientation_classifier which will be used to classify the orientation of crops (0, 90, 180, 270 degrees ccw).
We also provide a rectify_crops function in utils which will be used to rotate the crops accordingly.

Any feedback is welcome!

[codecov]

Codecov Report

Merging #721 (ec5c5ef) into main (2a18dc4) will decrease coverage by 0.12%.
The diff coverage is 94.62%.

@@            Coverage Diff             @@
##             main     #721      +/-   ##
==========================================
- Coverage   96.13%   96.01%   -0.13%     
==========================================
  Files         125      129       +4     
  Lines        4713     4792      +79     
==========================================
+ Hits         4531     4601      +70     
- Misses        182      191       +9     

Flag	Coverage Δ
unittests	96.01% <94.62%> (-0.13%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...octr/models/classification/magc_resnet/__init__.py	100.00% <ø> (ø)
doctr/models/classification/magc_resnet/pytorch.py	100.00% <ø> (ø)
...tr/models/classification/magc_resnet/tensorflow.py	100.00% <ø> (ø)
doctr/models/classification/mobilenet/__init__.py	100.00% <ø> (ø)
doctr/models/classification/predictor/__init__.py	100.00% <ø> (ø)
doctr/models/classification/resnet/__init__.py	100.00% <ø> (ø)
doctr/models/classification/resnet/pytorch.py	100.00% <ø> (ø)
doctr/models/classification/resnet/tensorflow.py	100.00% <ø> (ø)
doctr/models/classification/vgg/pytorch.py	100.00% <ø> (ø)
doctr/models/classification/vgg/tensorflow.py	100.00% <ø> (ø)
... and 21 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2a18dc4...ec5c5ef. Read the comment docs.

[fg-mindee]

Oh my bad, there is a misunderstanding: either we rename backbone into classification (which would make a lot of sense), or we add your modifications to "backbones", but no need to have both I think!

[charlesmindee]

We can merge both but we still need a predictor + model zoo for classification architectures, and we also need to have a cfg proper to this classification task (because the actual cfg of the mobilenet for instance is for a character classification, it is not the same input size, checkpoints, ...)
What do you suggest ?

[fg-mindee]

Thanks for the PR! I suggested a few changes!

[fg-mindee]

Thanks! I added a few comments, nothing major. Would you mind adding a corresponding section in the documentation please?

[fg-mindee]

Looks good, only missing the documentation 👌

[fg-mindee]

Looks good, thanks!