Expose data augmentation functions in config file about geo-deep-learning

[mpelchat04]

As of today, Albumentations has too many limitations (for our purpose), such as:

• Most of the transformations can only be done on 3 channel images. Our system is generic anough so someone would be able to use it with >3 channels input images. An example of this limitation is provided in the code here
• Some transformations assume that input images are uint8. Since we do not want to limit the use of GDL to uint8 images, I'm not sure we should use libraries that makes that assumptions.

Albumentations is not quite suited for our purposes. We'll reconsider the use of this library (or another one) to manage data augmentation later.

In the meantime, we'll still expose our data augmentation operations in the configuration file.