Bluecast

This release was all about code improvements. This included mainly:
* code refactoring
* bug fixes in the precise cv tuning

This release has new key features and bug fixes:
* Add a precise cv routine for overfitting-heavy routines.
* Add new preprocessing API that allows to do preprocessing on each fold during cross validation
* Enable the Shap values to also cover non-tree based custom models
* Fix the eval part to break when a custom model has been used due to missing experiment tracker implementation
* Give more design options for users of bivariate plots
* Add additional EDA plot for mutual information score
* Add additional user warnings
* Add additional logs
* updated docs

Add ROC AUC chart, predicted probas distribution & lift chart

Adding an optional grid search fine tuning capability on top of the usual autotuning. Also fix a few bugs around Xgboost parameters.

Version 0.40 adds an inbuilt experiment tracker. This works with the default Xgboost implementation. For custom models users need to call the tracker in the custom ml model definition.