[Fix] TF - add bf16 numpy dtype conversion

[felixT2K]

replace #1344

@odulcy-mindee you can check this later if you have a bit more time ^^

[codecov]

Codecov Report

Merging #1346 (1840dfd) into main (7374e89) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main    #1346   +/-   ##
=======================================
  Coverage   95.80%   95.80%           
=======================================
  Files         154      154           
  Lines        6906     6910    +4     
=======================================
+ Hits         6616     6620    +4     
  Misses        290      290           

Flag	Coverage Δ
unittests	95.80% <100.00%> (+<0.01%)	⬆️

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

Files	Coverage Δ
...etection/differentiable_binarization/tensorflow.py	95.16% <100.00%> (ø)
doctr/models/detection/linknet/tensorflow.py	94.89% <100.00%> (ø)
doctr/models/modules/transformer/pytorch.py	100.00% <100.00%> (ø)
doctr/models/recognition/crnn/pytorch.py	94.68% <100.00%> (ø)
doctr/models/recognition/crnn/tensorflow.py	97.59% <100.00%> (ø)
doctr/models/recognition/master/tensorflow.py	98.23% <100.00%> (+0.01%)	⬆️
doctr/models/recognition/parseq/tensorflow.py	97.34% <100.00%> (+0.01%)	⬆️
doctr/models/recognition/sar/tensorflow.py	98.56% <100.00%> (ø)
doctr/models/recognition/vitstr/tensorflow.py	97.61% <100.00%> (ø)
doctr/models/utils/pytorch.py	100.00% <100.00%> (ø)
... and 1 more