Trimming grid halo as part of the preprocessing

[jbusecke]

I just learned that some models for CMIP6 include the 'halo' cells (to e.g. accomodate the periodic boundary condition along the x axis) in the output, potentially causing issues with analysis and e.g. remapping to different coordinates.

If there is a quick automatic way to identify which cells (e.g. which end and how much) to trim, this would be a good thing to include in xMIP IMO.

[jbusecke]

I did some more digging here. I concluded from this that we really only need to remove two latitude/y columns at each side, IF they are all nans. But doing that naively might lead to trouble with e.g. datasets that have x and y swapped (#311).

Leaving this to simmer a bit, but if somebody needs a fix for now it could look like this:

• For both dimensions (x/y) Check if both ends have an all nan column, but not more than one row! This is to avoid clipping the y/latitude dimension, which most of the time has a bunch of nan rows over antarctica.

Still might need some more research on this, particularly with other output (e.g. land, atmos).