Implement neural network layers optimized for neuroevolution

[Talendar]

Feature:

Although TensorFlow is an awesome, high-performance framework, it wasn't designed with neuroevolution in mind. That said, perhaps NEvoPY should implement its own neural network layers, optimized for neuroevolution (without the usual "deep learning stuff", like optimizers and back-propagation). The idea is not to remove TensorFlow support from NEvoPY, but to have built-in neural layers that the user can opt to use.

Implementing a bug-free neural layer compatible with GPU is no trivial task, however. Also, I don't know enough about TensorFlow's core to judge if making that new implementations is worth. Some tests are needed in order to clarify this.