Cyclic-boosting

fix (in generic loss modes) for potential empty bins in multi-dimensional features

- enabled optional hierarchical iterations in training process
- added bin occupancy plots in training analysis plots of one-dimensional features
- added functionality to estimate full individual probability distributions from three estimated quantiles (e.g., from quantile regression) by means of quantile-parameterized distributions

fixed some misnomers

- some minor plotting fixes
- added quantile matching functionality to estimate full individual probability distributions from a few predicted quantiles

- added quantile regression and generic loss modes
- calculation of feature importances
- easier access to feature contributions to individual predictions (individual explainability)
- removed numba-scipy dependency (unblocking scipy upgrades)

- fix for publishing documentation
- use slightly newer scipy version