minimal fixes to run DataCollatorForWholeWordMask with return_tensors="np" and return_tensors="tf"

[dwyatte]

What does this PR do?

This PR addresses #13890 with the minimal fixes to run DataCollatorForWholeWordMask with return_tensors="np" and return_tensors="tf"

Specific problems addressed:

• Renamed np_call -> numpy_call
• Call _numpy_collate_batch instead of _tf_collate_batch when returning numpy tensors
• Fix size of random_words in numpy_mask_tokens
• Clone tensorflow tensors with tf.identity(tensor)
• Use tensorflow tensors’ built-in iteration instead of attempting to convert to list
• Change calls to tf.convert_to_tensor with dtype=tf.bool to tf.cast with dtype=tf.bool

I've only added a simple test to check for regressions vs all of the padded/unpadded cases as is done for DataCollatorForLanguageModeling

Fixes #13890

Before submitting

• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

CC @Rocketknight1 (looks like you did the initial Numpy/TF implementation of these data collators)

[dwyatte]

Bump @Rocketknight1 @LysandreJik @sgugger since transformers releases frequently and it would be nice to have this functionality.

This gives a vetted source for masked language modeling that is not https://github.com/google-research/bert/blob/master/create_pretraining_data.py or https://github.com/tensorflow/models/blob/master/official/nlp/data/create_pretraining_data.py in a package many people are going to be using for modeling (e.g., I could delete my code that is a copy/paste of the former and integrate against this instead for creating a large offline dataset).

[sgugger]

This looks good to me, thanks for the fixes!

[Rocketknight1]

Sorry for the delay, and thank you for the ping! This looks good to me. I've got one question about the convert_to_tensor versus cast call, but other than that this is a solid and necessary PR.