Bluecast

Adding additional model explainability:
* add SHAP waterfall charts plot
* add SHAP dependence plots
* enhance documentation with examples

This PR contains API changes alongside convenience and performance features:
* BlueCast & BlueCastRegression do not require the target column name during class instantiation anymore.
* t-SNE & PCA plots will scale input data automatically if not disabled by the user
* add `eta` param in grid search fine tuning
* raise `max_depth` in Xgboost param default
* add category encoding orchestration: If cardinality of a categorical column is below/equal a certain threshold, onehot encoding will be assigned, otherwise target encoding

The focus of this release has been EDA and explainability,
* add unit test for BlueCastCV in multiclass scenarios
* add calculation and print of out-of-fold scores for `fit_eval` in CV classes
* add new EDA visual: pairplot counts
* add new EDA visual: pie chart
* add new EDA visual: cumulative principal component variance

* add additional warnings in BlueCast checks
* add asserts into EDA functions so users receive better error messages
* add unit tests for additional checks
* update `univariate_plots` to also plot the target column if existent in the df
* update SHAP and numpy dependencies (SHAP versions of 0.39 and less required numpy 1.23 and less)
* enhance documentation to show how to use custom experiment tracking
* add data drift checks and qqplot for two samples
* update readme to show example plots

This release adds regression modelling. So far BlueCast supported binary & multiclass classification. With this version regression is supported as well. This covers the exact same functionality.

We also fixed minor bugs, updated tests and also the mutual information score to be usable for regression).

Small bug fixes for the precise tuning in multiclass tasks