Running version of PGGAN with Wasserstein loss

* Memory Leaks!!!
* No multi GPU support!!!

RefactoringDone

All models seem to work

Working CGAN with 3 losses

* tested with 5 landmarks & all landmarks
* tested on CelebA
* Images & features centered at 0, range [-1,1], loaded from numpy arrays

Tested CGAN implementation

* Landmarks should not be normalized -> Range: [0,1]
* Images should be normalized and centered -> Range: [-1,1]
Tested with CelebA64, 5 Landmarks, 100d RandomNoise and NO generated landmarks