get_image_spec needs a covariate file

[brenmous]

get_image_spec relies on having a covariate file to load image specifications from.

This isn't great as the user will always require a covariate file even if providing pre-intersected pickled data. We should change getting the image spec to during intersection and store this in the metadata of the model.

If we do on-the-fly reprojection/resampling/clipping in future it will be done there instead.