Separate ICs in zonal mean plots

[mcgibbon]

Currently the zonal_mean maps take the average across initial conditions, so the x-axis is forward step. It would be ideal if instead each initial condition were separately plotted across the y-axis of the zonal_mean plot - there's definitely space to do so. This would let us see the real evolution for each ensemble member.

If we add a zonal_mean aggregator for ensemble inference taking ensemble averages, we would still separate initial conditions across the y-axis, if there's more than one IC being ensemble'd.