feat(nlp): update PyTorch alt NLP to FastText

[athewsey]

Issue #, if available: #1

Description of changes:

• Update PyTorch alternative for NLP exercise to align to the FastText word embeddings now used in the TF.Keras version.
• Tweak eval batch size for faster compute
• Adjust loss metrics for better train/val consistency ('sum over batch' was actually taking mean)
• Align default instance type to g4dn.xlarge
• Tweak TF.Keras example to use same tokenize_*and*_pad_docs function name and use same numpy one-hot implementation as PyTorch - rather than relying on TF dependency.
• Remove unnecessary imports

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[yudho]

Tested successfully on these scenarios: Tensorflow Local Studio CPU kernel OK Tensorflow Sagemaker Studio CPU kernel OK PyTorch Local Studio CPU kernel OK PyTorch SageMaker Studio CPU kernel OK Tensorflow Local Notebook CPU instance OK Tensorflow Sagemaker Notebook CPU instance OK PyTorch Local Notebook CPU instance OK PyTorch SageMaker Notebook CPU instance OK

The prediction results are also reasonable.

The test does not include running it in local GPU environment for neither Studio nor Notebook.

The changes look good.