Deepvoxnet2

Not a small update actually :-)

Major changes:
- Subtle bugfix in the Buffer Transformer and added possibility for flip_probabilities="all" in Flip Transformer at n = 8.
- Introducing unet_generalized_v3: superior to v2, including bugfixes, bottleneck functionalities and simplifications.

Minor changes:
- Adding the pydot package to install_requires for plot_model functionality (please manually install graphviz also).
- Relaxing axis requirement in Mean Transformer and possibility to have same deformation in all axes in AffineDeformation when the option is specified by a single number.

Small updates:
- Default option chang in NormalizeMask Transformer (mean scale from 1 to 0).
- Some docstring changes
- Caching possibilities in some Modality classes.

In this release we updated setup.py to search for scikit-learn instead of sklearn and included nnUNetv2 dataset conversion from Mirc/Dataset classes, as well as introduced the shape attribute to the Modality class.

In this release there is extensive documentation of all components modules. This should make it easier for people to start using the library after having read through the tutorial notebook. Also, when using the .fit() method of the DvnModel class, now there is better functionality for using validation sets.

Woohoow! This is the first official release of the DeepVoxNet2 library. Each release now also comes as a PyPI package and a docker container on DockerHub.

First release of DeepVoxNet2.
This release includes a stable version of DeepVoxNet2, but tries out the actions pipeline of Github to trigger the creation of a PyPI package and a Docker container.