Improving the Arakawa & Konor domain

I've done a few things to improve the Arakawa & Konor test:

1. limited the perturbation to the middle of a wider domain
2. made the aspect ratio more atmosphere-like (domain is much wider than tall)
3. crucially, added cyclic boundaries

Single-layer radiation

The Lorenz computational mode is clear at the end of integration at t=172800s:

ExnerFoamCPinterpGrad still crashes after a while because the temperature constrast keeps on increasing. This is the result at t=16800s:

Two-layer initial anomaly

I've added sponges to the left, right and top of the domain to try and stop waves bouncing around. Again, the Lorenz computational mode is clear at t=172800s:

The Charney--Phillips model is getting too hot aloft but there is no computational mode:

With sensible domain size and periodic boundaries

Finally I'm getting towards a useful test case that looks similar on Lorenz and Charney--Phillips model variants:

I ran at dz and dz/2 resolutions. The Lorenz computational mode only appears with the grid-scale perturbation, not at dz/2 resolution.