RDA monthly average datasets with varying days in the month

[JohnLCaron]

Monthly average datasets using 1 day time unit (2) creates non-orthogonal, non regular time coordinates because number of days in the month varies.

could use calendar month unit, although this has not been accepted in CF.

datasets:
ds084.1
ds277.6 days 1
ds627.1.an_sfc hours
ds628.1.anl_mdl hours
ds628.3.fcst_mdl hours
ds628.5.fct_surf125_var_diurnal

these currently have wrong time, due to non-standard encoding:
ds094.2_t hours
ds094.2_dt

(partial list)

resol (6)
          28.0: count = 26
          29.0: count = 9
          30.0: count = 137
          31.0: count = 243
          59.0: count = 1
          61.0: count = 6

[JohnLCaron]

Need to fix this case so its not wrong. That is, make time unit = "day" or "hour" work.

Inclined to substitute calendar month, because its the simple thing to do.

Could have a "strict CF" mode when writing netcdf/CF. No sense complicating things because CF is so, um, conservative.

[JohnLCaron]

Issue #354:
See http:https://rda.ucar.edu/datasets/ds093.2/#!docs and click on "time_ranges.html" for information on how NCEP encoded metadata for monthly mean grids (which is not correct according to the GRIB2 standard)

Issue #353:
For CFSR - same as #354

[JohnLCaron]

Others have same problem, eg

ds628.1_anl_column (aka ds628.1/1)

and these currently barf because we are not yet handling non-orth time2D.

calendar month would solve problem. but first make non-orth (unique) work