[fitswcs] Units along spectral axes

[William Thompson]

I think that it's fair to make the statement that data providers using one of 
the non-linear algorithm codes of Table 2 cannot expect that the projection will 
be applied correctly unless the coordinate units are specified following the 
prescription of Paper I.  The best place for this statement would probably be in 
section 3.4.4, in the discussion of the RESTWAVa and RESTFRQa keywords, though 
it also applies to the projections involving air wavelengths, and the grism 
projections.