Region-based "onload"

We want users to be able to mark certain regions of the problem as requiring CPU (step sequential) processing. As a first step this can be a impl volume -> volume -> region lookup, but later if we have many different regions we could consider a BVH or something to accelerate the process.