Add Green's function method for potential field extrapolation

[wtbarnes]

The PFSS method of field extrapolation requires a Carrington map of the whole Sun as the lower boundary condition and is most useful for describing the large-scale structure of the global solar magnetic field. However, this is less useful if one is interested in the solar magnetic field at the scale of an active region. In these cases, it is often more useful to use the LOS magnetogram at a given instant in time as the lower boundary and extrapolate the field in a Cartesian box centered on that AR.

One such method for doing this is the Green's function method which is described in Sakurai (1982). I've done a bad thing and just created my own package to do this: https://github.com/wtbarnes/greensfield/ but I would really like to not do that.

In the long term, supporting these kinds of extrapolations will mean coming up with some kind of common data container for both kinds of extrapolations. However, I don't think this should prevent us from adding new kinds of extrapolations into this package.

That being said, the outputs of both kinds of extrapolations should have the same interface for tracing lines through the extrapolated volume. Currently, in greensfield, I've implemented an Extrapolator class that has a trace method on it, similar to Output. To do the tracing, it uses StreamTracer under the hood and returns a custom object that holds the coordinate and magnetic field line but there is no reason it couldn't make use of Fieldline and Fieldlines.

[wtbarnes]

Slightly more effort would be having a container for Cartesian extrapolations like this one and another container for spherical extrapolations. They could share a common lower-level container and the tracing interface could be implemented at that level.