Create a mask for a smaller image

[IFenton]

There is a suggestion that the image size we are using is larger than it needs to be, which is potentially limiting our ability to train more complex models. After analysis of the distance clouds are likely to move in the three hours of our prediction window, a smaller image size of 350km distance from the PVNet bounding box seems more logical.

Instead of having to re-download all the data, we will add the coordinates of this mask to the repo

[IFenton]

In the end, I did re-download all the data into Baskerville, it is in the eumetsat/cropped folder. I add the dimensions of this mask to the code in #78