Bluecast

* add effectivenes encoding into onehot encoder
* fix GPU detection
* Fix bugs when removing columns columns with zero variance or all nulls
* Add dynamic hyperparam tuning updates
* Fix logging of best params

Change overview:

* add wrappers for custom eval functions for single fold training runs (for CV and non-CV classes)
* strongly optimized hyperparameter tuning
* remove highly frequent GPU warnings messages
* enhance unit tests

This release contains multiple convenience improvements as well as a big performance upgrade.
- Rework logging
- fix datetime recognition in FeatureTypeDetector
- early stopping returns the best model now
- enable stratification for hyperparam tuning
- allow passing list with categorical columns to `BlueCastCV` instances
- add `train_config` option to decide on how many models to train in CV classes
- add option to tune hyperparameters multiple times via different random seeds
- updated docs

Taking various steps to increase the performance (especially for regression tasks):
* Set `min_child_weight` and `col_sample_by_level` to zero spans (min and max are equal) to speed up finding optimal spots for more important parameters. However users can still adjust these ranges on demand
* Fix range of `subsample` param which exceeded 1
* Remove adjusted losses
* Add multiple mean strategies to blend predictions of sub models in `BlueCastCVRegression`
* Updated poetry env
* Add optional sampling during training
* Enabled `use_full_data_for_final_model` by default

Better training performance:
* Updated hyperparameter tuning spaces
* Removed random seed tuning

* Update poetry lock to cover security updates of dependencies
* Reduce number of dependency plots shown by default from 5 to 1
* add conformal prediction for classification and regression
* add `predict_proba` methods for `BlueCast` and `BlueCastCV` classes
* add conformal prediction methods for `BlueCast` and `BlueCastCV`classes
* add string label handling to conformal prediction such as `BlueCast` & `BlueCastCV` return a set of labels rather than numeric encodings
* update how FeatureTypeDetector overwrites casted columns
* add new unit tests
* fix the `predict_proba` from multiclass tasks where sub model values aren't returned
* update the gpu support check to stop throwing the deprecation warnings for newer Xboost versions
* add eCDF function
* refresh and reorganize documentation
* add optional creation of Shap values during inference so users can plot the waterfall chart for any id/row from the new data