docs: performances update following up the crop splitting feature

[charlesmindee]

This PR adds the benchmark of all models with the new crop splitting feature activated, for some datasets it changes almost nothing, but for our private datasets (especially IDS with a lot of MRZ) it improves significantly performances and beyond that results are far more consistent when dealing with extremely long sequences, even if the exact match metric does not always reflect that (one small mistake over 45 chars is better than 45 mistakes...).

Any feedback is welcome!

[codecov]

Codecov Report

Merging #471 (fd7de48) into main (a735821) will decrease coverage by 0.02%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##             main     #471      +/-   ##
==========================================
- Coverage   95.79%   95.76%   -0.03%     
==========================================
  Files          95       95              
  Lines        3992     3992              
==========================================
- Hits         3824     3823       -1     
- Misses        168      169       +1     

Flag	Coverage Δ
unittests	95.76% <ø> (-0.03%)	⬇️

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

Impacted Files	Coverage Δ
...dels/detection/differentiable_binarization/base.py	91.35% <0.00%> (-0.62%)	⬇️

---

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 a735821...fd7de48. Read the comment docs.

[fg-mindee]

Looks good to me!