[CF-metadata] CF backwards compatibility with COARDS: identification of vertical coordinates by units of length

[Jonathan Gregory]

Dear John

For the sake of already existing data, I think we have to retain the
convention that identifies a vertical coord as having a positive attr or
units of pressure. Those are the rules given in CF 4.3.

The axis attribute is better, I agree, but it has always been optional
up to now, like many CF features. Hence we can't depend on it.

Best wishes

Jonathan