Fix of documentation after package reorganization

[tkornuta-nvidia]

Fixed several Sphinx-related issues that probably resulted from the reorganizations of a) the packages in general and b) of NLP collection in particular.

DONE:

• Cleaned up paths (EN/CN conf.py)
• Added 'transformers.tokenization_bert' to mocked up modules (EN/CN conf.py)
• Added 'transformers.tokenization_bert' to mocked up modules (EN/CN conf.py)
• Fixed several references to whole modules (EN/CN)
• Fixed several references to particular classes (EN/CN)

For example, .. automodule:: nemo.collections.nlp.huggingface.bert ->
.. automodule:: nemo.collections.nlp.nm.trainables.common.huggingface.bert

[lgtm-com]

This pull request introduces 1 alert when merging 09d4678 into 8d7bef6 - view on LGTM.com

new alerts:

• 1 for Wrong number of arguments in a class instantiation

[yzhang123]

import tokenizers as follows:
import nemo.collections.nlp as nemo_nlp
nemo_nlp.data.SentencePieceTokenizer

same for modules:

import nemo.collections.nlp as nemo_nlp
nemo_nlp.nm.[classifier]

pytorch_utils.py is gone after refactored

[lgtm-com]

This pull request introduces 1 alert when merging 8bf4389 into 8d7bef6 - view on LGTM.com

new alerts:

• 1 for Wrong number of arguments in a class instantiation

[lgtm-com]

This pull request introduces 1 alert when merging 9b5acd4 into 8d7bef6 - view on LGTM.com

new alerts:

• 1 for Wrong number of arguments in a class instantiation

[redoctopus]

Thanks for fixing :)