Fix the broken error detection in GFS_phys_time_vary.fv3.F90 and pointer issues in c3 scheme

[SamuelTrahanNOAA]

There were several bugs in the error detection in GFS_phys_time_vary.fv3.F90. Errors might go undetected, or may be reported incorrectly.

1. Fixes ozinterp and h2ointerp data is read when it is not needed #108
2. Fixes The noahmptable.tbl file is mandatory, even for configurations that don't need it. #107
3. Fixes GFS_phys_time_vary_init does not report errmsg/errflg correctly due to thread race condition #105
4. Fixes incorrect filename in warning message for missing noahmptable.tbl #104
5. Adds errflg=0 and errmsg='' in two routines that were missing them.
6. Fixes FV3_HRRR_c3 suite crashes in c3 calls on Cheyenne with gnu compiler in debug mode. #101

[grantfirl]

Thanks, @SamuelTrahanNOAA. So, if I'm understanding correctly, this problem/solution should be general for any non-run CCPP phase that has code setting errmsg/errflg within an openMP section. Do you know if there is any other code (e.g. in init/timestep_init) phases that are doing this that we need to fix, or is the problem confined to this code as far as you know?

[SamuelTrahanNOAA]

I don't see any other OpenMP sections used in initialization. Perhaps I missed something in my grepping? As far as I know, no such bug remains.

And, yes, this is completely general.

[dustinswales]

@SamuelTrahanNOAA This looks great to me.

[SamuelTrahanNOAA]

I've merged this PR into #102 since the UFS code managers plan to merge the two simultaneously.