[CF-metadata] Variable Associations (was CF convention for vector quantities)

[Jonathan Gregory]

Dear Bryan

> I think it's quite clear that the vertical components don't fit into the 
> algorithm you describe.

No, they don't. I was describing an algorithm to generate the file for Bert's
2D vector case. The vertical components aren't named so systematically.

I do think that knowledge of higher-level associations is useful, but there
are many of them, they are application-dependent, and they cannot all be
foreseen at the time of writing the file. Hence I think it's better to put
them in external metadata. The role of CF metadata is to allow you to identify
and describe the quantity, not to prescribe what to do with it.

Cheers

Jonathan