Support fine-tuning models

[leewyang]

This PR adds support for fine-tuning (aka. "continued training" or "incremental learning") of existing pre-trained models on new datasets without having to retrain the entire model from scratch using all of the original training datasets. This can be useful if the original data is unavailable, or if the new data is customer-specific and the fine-tuned model is only intended for that customer. Note that the alternative is to train a new model from only the new data, but this requires a "sufficient" number of pairs of CPU/GPU applications and sqlIDs.

Changes

1. Updated qualx README.md with fine-tuning instructions.
2. Added a --base_model argument to the train CLIs.
3. Added xgb_model.save_config() calls to export the training hyper-parameters while saving the trained models.
4. Added new model.json.cfg files from training.
5. Load base_model and cfg when training.
6. Allow for multiple train/test spit functions, including dataset-specific split functions in plugins.
7. Fix bug in get_dataset_platforms() to allow for any paths (vs. local relative).

Test

Following CMDs have been tested:

External Usage:

spark-rapids train --base_model

Internal Usage:

python qualx_main.py preprocess
python qualx_main.py train
python qualx_main.py train --base_model
python qualx_main.py evaluate
python qualx_main.py compare

[eordentlich]

👍

[mattahrens]

Updated PR description notes to be clear that new argument is --base_model and not --base-model to follow tools conventions for arguments.