Things that Could be Improved in This Thorn
===========================================

This thorn works with non-unigrid Cactus, but it doesn't work very
efficiently: it does the 2D elliptic setup and solve all over again
for each local Cactus (3D) grid.  It would be more efficient to move the
2D elliptic stuff out into a separately scheduled routine which would
be done once (for Carpet, it would be scheduled in level or global
mode), and have only the interpolation to the 3D grid done for each
local Cactus grid.

Part of this same change might also be to store the 2D elliptic results
in Cactus grid arrays rather than native Fortran 90 arrays.  This would
let us use standard Cactus I/O methods to output the 2D elliptic results
instead of the hard-wired code we use right now.