NaN values in some input files cause problems with ifx compiler

[billsacks]

Brief summary of bug

As found by @jedwards4b - SMS_D.ne30pg3_ne30pg3_mt232.FHISTC_LTso.derecho_intel is failing when updating to ccs_config_cesm1.0.75, which switches to the ifx compiler, due to some input files having NaN values. Based on consultation with @ekluzek we should change these files to have no NaN values.

General bug information

CTSM version you are using: ctsm5.4.003

Does this bug cause significantly incorrect results in the model's science? Unknown, but currently believed to be No.

Configurations affected: Unknown

Details of bug

From @jedwards4b -

I was looking at

and it's crashing in dshr_strdata_mod.F90 on this line:
if (stream%mapalgo == 'none' .and. any(data_dbl1d == fillvalue_r8)) then

So I looked at the data in the file that it's trying to read and the file has NaN as a fill value:

double p_ac_TBD(time, lat, lon) ;
p_ac_TBD:_FillValue = NaN ;
p_ac_TBD:long_name = "Air-conditioning penetration rate for Tall Building District (TBD) class" ;
p_ac_TBD:prediction_method = "Assume all 1; filled" ;
p_ac_TBD:units = "-" ;

I then tried a run where I convert the NaN _FillValues to 1.e38 it got past that file and failed in a later
stream.

and then:

The next one is even trickier because the _FillValue is
double Z0a(time, lat, lon) ;
Z0a:units = "cm" ;
Z0a:_FillValue = 1.e+32 ;
Z0a:long_name = "Prigent et al. (2005) 0.25x0.25 aeolian roughness length in cm" ;

But if you ncdump the file you see that the field is full of NaN values anyway - probably over ocean points?

Important details of your setup / configuration so we can reproduce the bug

Testing was done in the context of CESM at 68fe606, which uses ccs_config_cesm1.0.75.

Also pinging @briandobbins on this.