Fix precision for writing field attributes

[xylar]

It is necessary that the precision of floating-point attributes match the precision of the field they are associated with. In particular, when the _FillValue has a different precision from the field it is associated with, PIO complains and will not add the attribute.

Fixes #5235
[BFB]

[xylar]

Testing a CMPASO-TL319_EC30to60E2r2 run on Cori. I'll take this out of draft mode if I find out that that run is successful.

[xylar]

Testing

The fix does seem to have worked as expected. See results in:

/lcrc/group/e3sm/ac.xasay-davis/scratch/chrys/20221026.CMPASO-JRA1p4_TL319_EC30to60E2r2.fix-time-series-stats.chrysalis/run

a 1-month MPAS-O simulation in which timeSeriesStatsMonthly shows the expected single-precision fill values:

$ ncdump -h 20221026.CMPASO-JRA1p4_TL319_EC30to60E2r2.fix-time-series-stats.chrysalis.mpaso.hist.am.timeSeriesStatsMonthly.0001-01-01.nc | grep "_FillValue"
		timeMonthly_avg_layerThickness:_FillValue = 9.96921e+36f ;
		timeMonthly_avg_normalVelocity:_FillValue = 9.96921e+36f ;
		timeMonthly_avg_activeTracers_temperature:_FillValue = 9.96921e+36f ;
		timeMonthly_avg_activeTracers_salinity:_FillValue = 9.96921e+36f ;

However, I am not seeing the expected fill values in the fields themselves, e.g.:

 ncdump -v timeMonthly_avg_activeTracers_temperature 20221026.CMPASO-JRA1p4_TL319_EC30to60E2r2.fix-time-series-stats.chrysalis.mpaso.hist.am.timeSeriesStatsMonthly.0001-01-01.nc | tail
  27.55785, 27.53584, 27.52425, 27.51439, 27.50551, 27.49708, 27.38161, 
    26.46217, 24.78305, 23.25717, 21.93985, 20.68438, 19.45276, 18.26175, 
    17.16108, 16.16831, 15.26815, 14.42972, 13.62875, 12.85962, 12.13178, 
    11.46993, 10.88002, 10.35534, 9.884884, 9.467792, 9.11305, 8.818574, 
    8.540646, 8.274009, 8.000038, 7.712546, 7.409822, 7.091074, 6.753331, 
    6.39432, 6.000737, 5.534378, 5.060202, 4.568097, 4.028841, 3.540712, 
    3.114681, 2.73235, 2.406448, 2.104124, 1.880408, 1.692876, 1.552158, 
    1.462184, 1.408809, 1.356857, 1.314016, -1e+34, -1e+34, -1e+34, -1e+34, 
    -1e+34, -1e+34, -1e+34 ;

This suggests that work is still needed in timeSeriesStats to make sure that fill masks from the original fields are also used for the time-averaged fields. Since this is a different bug, I won't try to address it here.

[xylar]

@cbegeman, could you take a look at this when you can? If you want to run a test, that's fine. If you prefer to look at my test results, that's okay, too.

[cbegeman]

@xylar This looks great. Thanks for tackling this. I'll get to the related issue #5253 soon.

[jonbob]

passes sanity testing, merged to next

[jonbob]

merged to master

[xylar]

Thanks @jonbob and @cbegeman!