Use knowledge data for phase 1 training and skills data for phase 2 training in standalone script

[MichaelClifford]

Closes #86

• Updates the data_processing_op() to output two datasets,skills_processed_data and knowledge_processed_data instead a single processed_data output. These new datasets are stored in .../processed_data/knowledge and .../processed_data/skills respectively.

• This changes requires that the SDG data used contains both a knowledge_train_msgs*.jsonl file as well as a skills_train_msgs*.jsonl file (which it should, but may cause a failure if a small PR is used to generate the SDG data).

• Updates the PYTORCH_TRAINING_JOB template in standalone.tpl to use the knowledge_processed_data for phase 1 training and skills_processed_data for phase 2.

Note: This PR does not add this feature to the KFP, I need to update the pytorchJob_manifest_op() to reflect this change. However, that function has fallen a bit behind the standalone version at this point so will bring it back to parity in a later PR.

[sallyom]

this code looks clean, let's merge it - with the caveat that it hasn't been tested in the pipeline so there may be a follow-up PR to fix the pipeline - so long as this has been tested and proved to work with standalone script, I'm good with merging this!
LGTM