Better document the 2 extra dimensions to our 2d/3d convolution

In the Theano conv tutorial, we don't talk about them:

http://deeplearning.net/software/theano/tutorial/conv_arithmetic.html

But this is explained in the original conv explanation arxiv paper in or around figure 1.3. We should move that/adapt that to Theano conv tutorial and make sure it explain the 2 dimensions.

https://arxiv.org/abs/1603.07285