Fix array allocation problem with new nfix tests

[slevis-lmwg]

@glemieux got segfault errors in a bunch of izumi nag tests on the tmp-241219 branch:

FAIL ERP_D_Ld5_P48x1.f10_f10_mg37.I1850Clm50Bgc.izumi_nag.clm-ciso RUN time=223
FAIL ERP_D_Ld5_P48x1.f10_f10_mg37.I1850Clm60Bgc.izumi_nag.clm-ciso RUN time=299
FAIL ERP_D_Ld5_P48x1.f10_f10_mg37.I1850Clm60Bgc.izumi_nag.clm-ciso--clm-matrixcnOn RUN time=260
FAIL SMS_D.f10_f10_mg37.I2000Clm60BgcCrop.izumi_nag.clm-crop RUN time=38
FAIL SMS_D_Ld1_P48x1.f10_f10_mg37.I2000Clm45BgcCrop.izumi_nag.clm-oldhyd RUN time=48
FAIL SMS_D_Ld5.f10_f10_mg37.I2000Clm50BgcCrop.izumi_nag.clm-irrig_alternate RUN time=50
FAIL SMS_D_Ld65.f10_f10_mg37.I2000Clm60BgcCrop.izumi_nag.clm-FireLi2024GSWP RUN time=49
FAIL SMS_D_P48x1_Ld5.f10_f10_mg37.I2000Clm50BgcCrop.izumi_nag.clm-irrig_spunup RUN time=53
FAIL SMS_Ld5_D_P48x1.f10_f10_mg37.IHistClm50Bgc.izumi_nag.clm-monthly RUN time=128

@glemieux confirmed that the same tests pass in ctsm5.3.016, so we suspect that the bug entered in the first tmp-241219 branch tag, i.e. when @slevis-lmwg merged b4b-dev (ultimately #2917).
@slevis-lmwg ignored the izumi tests when merging b4b-dev because izumi had just been upgraded and was giving problems, including not having room for new baselines.

@slevis-lmwg will work on debugging this starting today. The fix needs to go on b4b-dev as well.

[slevis-lmwg]

I started with one of the Clm6 SMS tests from the list:
SMS_D_Ld65.f10_f10_mg37.I2000Clm60BgcCrop.izumi_nag.clm-FireLi2024GSWP.

[glemieux]

For future reference, here's the location of the results from the tests I ran:

tmp241219 results: /scratch/cluster/glemieux/ctsm-tests/tests_0103-202929iz
ctsm5.3.016 results: /scratch/cluster/glemieux/ctsm-tests/tests_0103-164806iz

[slevis-lmwg]

Promising news:
The SMS test that I started with, gets past the segfault error and appears to run smoothly now.
I have now submitted another SMS and one ERP.

[slevis-lmwg]

The second SMS test and the ERP test also passed. It probably makes sense for me to open a PR and run the test-suites next.

[ekluzek]

We talked about this at the CTSM SE meeting this morning. Our notes from there:

• Good to do to figure out how to prevent in the future
• It didn’t come in in the original implementation – good as most work will be fine with it
• It came in with some “b4b” refactoring
• The issue means that using the new method would be wrong in some fashion, e.g. in 2-deg I-case spinup and historical baseline for PPE with BNF tag (ctsm53017_f19_BNF_[AD, SASU, pSASU, hist]) NCAR/LMWG_dev#88
• Note that nag caught this even though test exercising the option didn’t exist (because parameters were being read and arrays allocated even when option was off)
• tmp-241219.n03.ctsm5.3.016 will be adding one izumi and one derecho test
• Takeaways:
• Make sure there is at least one test for new namelist options coming in.
• When refactoring something make sure it’s exercised in tests and give identical results

@briandobbins this is the issue with the problem we discussed in CSEG. The fix is small and came in with tmp-241219.n03.ctsm5.3.019. The tag WITH the problem is: tmp-241219.n01.ctsm5.3.019 so it would be good to try compiling with more aggressive checking options in that tag.