Move texar to extra require

[hepengfe]

This PR fixes #719.

Description of changes

• raise ImportError with suggestions when the extra library texar-pytorch is not installed

Possible influences of this PR.

• User doesn't need to install pytorch by default for using main texar functions

Test Conducted

• Implemented texar_non_dependency_test.py that imports classes that doesn't have dependency on texar-pytorch but might trigger importing torch accidentally through __init__.py or importing torch at the head of class files.

[hunterhector]

@feipenghe @mylibrar Please notice that the solution cannot be simply throwing exceptions at run time.

If we decide Texar would be extra required, we need to provide a good extra requirement install option in setup.py, and in the documentation here: https:/asyml/forte#download-and-installation, explain which ones require which extra requirement.

Note that you also cannot simply say pip install forte[texar], it is not intuitive. It should be related to the function itself. For example, if the users need to use the models (srl, ner), the extra requirement should be pip install forte[forte-models]. If the function is related to data augmentation models, the extra requirement could be pip install forte[data-aug-models].

Here is the goal: the user shouldn't encounter any exception if he follows the documentation. The exceptions are pointers for a user to notice the documentation, not documentations by themselves.

[hunterhector]

why don't we run these steps before the Install Texar or even the Install deep learning frameworks step?

[hepengfe]

Since we still have dependencies on other libraries in the dependency matrix. In this PR, we only need to remove texar. In the later PR that we set up a dependency env for different modules, these lines of code will be removed.