[GSProcessing] BERT Tokenizer

[jalencato]

Issue #, if available:

Description of changes:

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

---

Beside implementing the BERT Tokenizer feature

• Add etype_featsize dict in graph loader to fix a bug
• Add dependency for setuptools that is necessary for emr-serverless run
• Reformat the _process_node_feature and _process_edge feature part to process tokenize feature and also allow enough backward compatibilities.

Result:

• Constructing full mag dataset with tokenize feature for 111 minutes. Refer to the example here: https://github.com/awslabs/graphstorm/blob/main/examples/mag/mag_v0.2.json

[thvasilo]

Overall the implementation looks solid. Let's work on some changes before merging.

[thvasilo]

Small changes requested.

[thvasilo]

A few changes left.

Let's ensure the following:

• run black on the codebase to ensure formatting is correct.
• Increment the version in pyproject.toml to 0.2.2.
• Ensure both images build and can run a text transformation using the new build argument on both EMR-S and SM.

[thvasilo]

LGTM, we should run some benchmarks to see if we have any bottlenecks with the implementation.

[jalencato]

LGTM, we should run some benchmarks to see if we have any bottlenecks with the implementation.

Update our result on the mag with the tokenized feature.