Halo/perimeter cells

[MattEdwards94]

At the edge of the domain, 'cells' do not currently exist. This means that they could not be split if it was deemed necessary at this location.

Implement Halo cells which contain 'fake' control points outside of the image domain.

This raises a couple of issues though:

• Splitting a 'halo' cell near the perimeter may not result in an additional sample being placed within the border. How should this be handled? Should the cell be recursively split until a new sample is added?

• How should the positioning of the grid be decided? Should the grid be shifted to ensure roughly even distance to image domain from the 'outside' grid points? How would this consider masks?
  

• Perhaps we have an initial refinement algorithm which reduces the distance from an interface (mask or image boundary) to the nearest sample. How would this be controlled though? Would it allow sampling all the way down to pixel level? Perhaps down to within some distance?