Use dataloader_drop_last in TF dataset

[setu4993]

Follows PR #4757, fixes issue #4891. I didn't realize then that TFTrainingArguments inherited TrainingArguments and all of the same options would be available there as well.

Adding this to get to feature parity across PyTorch and TF trainers.

[codecov]

Codecov Report

Merging #4925 into master will increase coverage by 0.07%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #4925      +/-   ##
==========================================
+ Coverage   77.10%   77.17%   +0.07%     
==========================================
  Files         128      128              
  Lines       21723    21723              
==========================================
+ Hits        16749    16765      +16     
+ Misses       4974     4958      -16     

Impacted Files	Coverage Δ
src/transformers/trainer_tf.py	19.04% <ø> (ø)
src/transformers/file_utils.py	73.79% <0.00%> (ø)
src/transformers/modeling_utils.py	90.61% <0.00%> (+0.11%)	⬆️
src/transformers/trainer.py	39.48% <0.00%> (+0.23%)	⬆️
src/transformers/modeling_xlnet.py	78.65% <0.00%> (+2.29%)	⬆️

---

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 e80d6c6...a9099e0. Read the comment docs.

[julien-c]

LGTM from a quick glance but I'll let @jplu chime in

[jplu]

Nice! LGTM!!