[CF-metadata] CF and a representation of probalistic forecasts

[Jonathan Gregory]

Dear Jamie

If the different ways of calculating realization_weights produce distinct
scientific quantities, as I suppose they do, then they should have
different standard_names, so you can tell them apart. I think that more than
one set could be pointed to as ancillary variables nonetheless. In recording
a statistic in cell_methods evaluated across realizations, you might add a
note in parethensis about which weights were used. If it became a frequent
requirement to distinguish between options, a standard way to indicate this
in cell_methods could be proposed.

Cheers

Jonathan