We want to be able to produce plots in GPU runs. This was partially implemented in a branch that wasn't merged into main (see here). It's accomplished by making a new space on the CPU, then copying fields from GPU to CPU. This will be slow and allocate a lot, but might still be worth it for debugging and postprocessing.

Maybe we should add a flag to optionally enable this in GPU runs, to avoid accidentally converting all the fields for plotting and slowing down GPU runs

Update: All plotting in the driver is currently controlled by 3 flags, energy_check, diagnostics, and ci_plots, so they can all easily be turned off (energy_check and ci_plots are off by default)

part of #571