[refactor][fix]: Part1 from use datasets for recognition task

[felixdittrich92]

@charlesmindee @frgfm
Not much to review in this PR only a view slight modifications & preparations for the following PR 🤗

modifications:

• move extract_crops and extract_rcrops to geometry for reuse (otherwise fails with circular imports)
• fix 2 dataset boxes
• add tensor_from_numpy
• add minor additional tests for dataset/utils

Link to PoC: #867

[codecov]

Codecov Report

Merging #889 (7ddb39a) into main (2c697ff) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #889      +/-   ##
==========================================
+ Coverage   94.92%   94.93%   +0.01%     
==========================================
  Files         133      133              
  Lines        5280     5293      +13     
==========================================
+ Hits         5012     5025      +13     
  Misses        268      268              

Flag	Coverage Δ
unittests	94.93% <100.00%> (+0.01%)	⬆️

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

Impacted Files	Coverage Δ
doctr/datasets/datasets/pytorch.py	100.00% <100.00%> (ø)
doctr/datasets/datasets/tensorflow.py	100.00% <100.00%> (ø)
doctr/datasets/utils.py	95.83% <100.00%> (+0.59%)	⬆️
doctr/models/_utils.py	96.00% <100.00%> (-1.60%)	⬇️
doctr/models/predictor/base.py	82.08% <100.00%> (+0.27%)	⬆️
doctr/utils/geometry.py	100.00% <100.00%> (ø)

---

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 2c697ff...7ddb39a. Read the comment docs.

[felixdittrich92]

Will add a missing test tommorow

[fharper]

@charlesmindee can you review this PR please? :)

[charlesmindee]

Thanks for this first step, it is indeed a good idea to use datasets for both tasks