Expectations/Capabilities related to writing output

[kbarnhart]

@mdpiper @mcflugen (this might be more appropriate over in the bmi repo. But I think it is better here.

I'm writing the terrainbento bmi (in progress PR here).

When I designed terrainbento I thought about how to make output easy to use. Each saved timestep writes output in a netcdf and then I have a function that makes it easy to compile these into an Xarray dataset.

But if I want to run one of terrainbento's 28 models coupled to something else with PyMT I probably only want to save the state variables once. And I probably want to save all the variables in the same Xarray dataset.

Is this possible? Are there things I need to do with my BMI to make this work?