Extend NetCDF zlib compression to UGRID

[bjlittle]

✨ Feature Request

Reference #6552.

NetCDF zlib compression has been extended to cover auxiliary coordinates and ancillary variables with the same shape as the cube data payload (excluding string dtypes).

@pp-mo Had thoughts on offering a richer, more flexible opt-in experience for users, which may be part of the scope of this task (see #6552), but primarily the focus would be to also support zlib compression of associated UGRID variables.

This would be a major win for the community with regards to minimising the serialized NetCDF footprint on disk, which ATM is a common objection.

[pp-mo]

@pp-mo Had thoughts

To be definite, I was mostly speculating whether it might be more flexible and more easily managed if we used context managers instead of keyword controls to control saving aspects generally.
This turned out well for controlling chunking on loading, since it allows the controls to be targeted at specific file variables, or dimensions, which don't relate directly to the Iris objects.
So broadly, that seems like it might apply here too.

By taking control management outside the per-cube context of the save operation, this has some other possible (non-functional) benefits too

• (a) can be managed and documented as a separate thing, and
• (b) does not add to the parameter list of all intermediate routines.

Unfortunately though, this approach isn't so simple to apply to saving as loading, because the user will need to "predict" which variables will be created with what var-names. And in some cases that isn't obvious -- like bounds variables or duplicate names -- so it will require more explanation.