Tabml

Change logs:

* Fix inheritance order in housing custom model wrapper
* Add R2 metrics

Change logs:

* Simplify custom model wrapper API, now it can be specified in pipeline config
* Move tests and examples folder to tabml folder
* Use safe_load in yaml parser
* Code cleanup

Small fixes:

* Replace config paths ending with `.pb` by `.yaml`

Change logs contain mostly [miscs](https://github.com/tiepvupsu/tabml/issues?q=label%3Av0.1.13+):

* clean up unused function
* Replace os.path by pathlib.Path
* Explicitly use params instead of config in main components
* Load datasets if they exist in local, download otherwise
* Move load_transformers to ModelInference initialization in inference.py

Change logs:

* Merge Trainer to ModelWrapper
* Migrate all configs to yaml, configs are now validated by [pydantic](https://pydantic-docs.helpmanual.io/). This way users have more flexibility to defined parameters.
* Miscs:
- Allow users to specify `transformers.pickle` path.
- Clean unused codes and packages

Change logs:

* Support xgboost and catboost
* integrated with SHAP
* remove old feature_importance (now replaced by SHAP feature_importance)
* allow analyzing model on a subset of training data
* miscs: clean up test datasets, reuse .isort.cfg, log model_type in mlflow