Update io_hdf5 for HDF5 1.14

[mathomp4]

This PR updates io_hdf5.F90 to support HDF5 1.14. Testing showed that between HDF5 1.10 (in current baselibs) and HDF5 1.14 (in GEOSgcm v12 Baselibs) something changed in how you need to open/close/etc. HDF5 files.

In my limited testing (Intel conus and global), this PR seems to be zero-diff with the current Baselibs and as such, can be dribbled in now before updating Baselibs.

I'll keep draft until a full suite test can be made.

[mathomp4]

@weiyuan-jiang has kindly undertaken a full suite test for this.

[mathomp4]

Well, huh. @weiyuan-jiang did a run for me and:

Full Log File: /discover/nobackup/wjiang/SystemTests/logs/LDAS/2025-08-20/log.12:42:05

Runtype         Clone Build Build Time Model Run/Compare Assim Run/Compare
-------         ----- ----- ---------- ----------------- -----------------
conus           pass  pass  12 min     pass/pass          -- / --
global           --    --    --        pass/pass         pass/pass
globalcs         --    --    --        pass/pass         pass/pass
globalcnclm4     --    --    --        pass/pass          -- / --
debugconus       --   pass   6 min     pass/pass          -- / --
aggconus         --   pass  12 min     pass/FAIL          -- / --
aggglobal        --    --    --        pass/FAIL         pass/FAIL
aggglobalcs      --    --    --        pass/FAIL         pass/FAIL
aggglobalcnclm4  --    --    --        pass/FAIL          -- / --
gnuconus        pass  pass  28 min     pass/pass          -- / --
gnuglobal        --    --    --        pass/pass         pass/pass
gnuglobalcs      --    --    --        pass/pass         pass/pass
gnuglobalcnclm4  --    --    --        pass/pass          -- / --
gnudebugconus    --   pass  17 min     pass/pass          -- / --

The fact that a CONUS run changed makes me think it's something else instead of this PR?

[gmao-rreichle]

The fact that a CONUS run changed makes me think it's something else instead of this PR?

@mathomp4 : The non-0-diff results are for "agg". I doubt they matter. I can spot-check to confirm they're the usual roundoff. Was the test run using the new baselines plus the new h5? Or just the new h5?

[mathomp4]

The fact that a CONUS run changed makes me think it's something else instead of this PR?

@mathomp4 : The non-0-diff results are for "agg". I doubt they matter. I can spot-check to confirm they're the usual roundoff. Was the test run using the new baselines plus the new h5? Or just the new h5?

@gmao-rreichle It should just be the hdf5 updates, see GEOS-ESM/GEOSldas#833 which I guess I can close now...or rather should have not opened? I'm new to GEOSldas GC PRs. :)

I am just now in the process of issuing new Baselibs, etc. but that'll be a longer term process since I need to build them up in places.

[gmao-rreichle]

@mathomp4 : Sorry for the silly question about h5 vs. new baselibs, I just looked it up myself. I'm about to have a few meetings, I'll check the diffs as soon as I can. It's a bit odd that the code changes in io_hdf5.F90 would create non-0-diff results even for "agg", but if the diffs are roundoff (TBC), I think it'll be ok to merge

[mathomp4]

@mathomp4 : Sorry for the silly question about h5 vs. new baselibs, I just looked it up myself. I'm about to have a few meetings, I'll check the diffs as soon as I can. It's a bit odd that the code changes in io_hdf5.F90 would create non-0-diff results even for "agg", but if the diffs are roundoff (TBC), I think it'll be ok to merge

Oh, not silly. I sort of had to go check to make sure myself. I was doing a lot of testing and I really don't want some temporary baselibs getting into a model :)

[gmao-rreichle]

Mystery solved... The "agg" comparisons fail only for the "ldas_catparam" file. This file contains time-invariant model parameters and is generated in pre-processing. For the most part, the parameters are read from somewhere and immediately written out into the "ldas_catparam" file without any calculations. But one field ("cdcr1") involves a calculation. There's a roundoff diff in this calculation when the test is run on Cascade Lake and the baseline was run on Milan. I'll go ahead and merge the PR.