CoordConv and spatial resolution

[mpelchat04]

CoordConv works by giving convolution access to its own input coordinates through the use of extra coordinate channels Liu et al., 2018.

With remote sensing imagery, we also have access to the spatial resolution of the raster, which could be included as an input layer. Including prior knowledge into the model should improve generalization when training on multiple spatial resolution datasets.

[mpelchat04]

Closed in #109