Store coil fields for GS solve

[bclyons12]

Store the poloidal flux from each (real) PF coil in separate fields, so that they don't need to be recomputed at each time step. This has some awkwardness/hacky coding with respect to the coil currents, since we only save one poloidal field per physical coil, but still save a current per filament, so it needs to get summed up. That could be simplified, but this works as is.

Timing Test for GS solver with resistive wall & PF coils

DIII-D case (18 coils)

Old method: ~168 s
New method: ~100 s [9 s to store, 91 s to iterate]

KSTAR case (74 coils)

Old method: ~191 s
New method: ~87 s [21 s to store, 66 s to iterate]

[bclyons12]

@nferraro This currently works and the RMP regression test with igs_store_coils=1 passes, but it's much slower than the old method for a large number of coils. Any idea why? Too much memory access? My instinct is that it should be much faster than having to compute the Green's function for every coil at every point in space each GS iteration.

[bclyons12]

b2a8245 sped things up by ~33%, but still much too slow

[bclyons12]

It's because I'm storing a field for each filament, which is a very large number. I'm working on a change to store one per real coil.

[bclyons12]

db5e684 makes the GS solver faster than the original. See top comment for more info.

[bclyons12]

@sjardin You may be interested in this.