Extend convolution layer to support padding

[accosmin]

References:

• http://deeplearning.net/software/theano/tutorial/conv_arithmetic.html
• https://adeshpande3.github.io/A-Beginner%27s-Guide-To-Understanding-Convolutional-Neural-Networks-Part-2/
• http://cs231n.github.io/convolutional-networks/

[accosmin]

It's not actually needed because we can construct and generalize Residual Networks without padding.