Implements #348 support for groundtruth

It does not currently use a 3D model directly, this is expected to be done using a Blender script to output test data. It does allow for static image FTL files with a ground truth channels and a way to display a middlebury evaluation using the ground truth channel.

Further work: once multi view test data is available, it must be possible to render a virtual view twice with our depth and the ground truth. We may then assess both depth error and colour reprojection error.