Enhance augmentation on tub data:

* Add image augmentation as a standalone Tub member function which replaces the images in place.
* Refactor the augmentation to separate into conversion between numpy arrays and PIL images and the augmentation functions
* Add a donkey command 'donkey tubaugment' to augment tubs in place or clone the tub into a now one with augmentation and keep the original.
* Add a test that checks the augmented data only differs slightly from the original images