Use ACCESS-NRI fork of generic tracers with accessom_coupler

[dougiesquire]

See mom-ocean#388 for context and details.

This PR was originally made to mom-ocean/MOM5 but it was decided that we would make the changes here first while some details about the build system and testing are still being resolved. Comments from @aidanheerdegen's review of the original PR have been incorporated here.

Note, this PR includes the ACCESS-NRI fork of generic_tracers as a git submodule.

Note also, this PR removes the need for the ACCESS-OM-BGC build type (and hence the CSIRO_BGC pp def). The ACCESS-OM build type now always builds with the USE_OCEAN_BGC pp def and generic tracer WOMBAT can be configured at runtime. I’ve verified that always building with USE_OCEAN_BGC doesn’t impact performance when generic tracers are not configured. To do this, I ran two ACCESS-OM2 experiments on the NCI Gadi supercomputer:

• Exp 1 uses this configuration modified to run for 1 year at a time and to use a MOM5 executable built from 59e0aad (the most recent commit on main at the time of opening this PR).
• Exp 2 uses the same configuration, but the MOM5 executable used includes the changes in this PR.

I ran three years of each. The experiments were run at the same time. The chksum outputs in the logs are identical between the experiments and the walltimes reported by PBS for each year are given in the table below.

	Exp 1	Exp 2
year 1	959 s	908 s
year 2	1031 s	1002 s
year 3	1148 s	1028 s
average	1046 s	979 s

Closes mom-ocean#388

[dougiesquire]

Build CI is failing. I haven't looked in any detail, but one thing sprang to mind - does the build CI need me to make this from a branch in this repo (rather than a fork)? @CodeGat

[CodeGat]

Hi @dougiesquire - I reran the CI to see if the error was reproducible, and yeah I think you're on the money. Seems like it's looking for the ref on ACCESS-NRI/MOM5 rather than your fork, since that is what is defined in the SPD.

[dougiesquire]

Thanks @CodeGat . I'll also need to add submodules = True to the SPD. Does the build CI just use the latest version on main of ACCESS-NRI/spack-packages?

[CodeGat]

No, it uses the latest tagged version - in which case you'll need to push a tag after you've merged the PR (2024.08.08). I'll have to do a workaround after that because this issue is still plaguing me :D ACCESS-NRI/build-ci#170

[dougiesquire]

Closing this and will reopen an equivalent PR from a branch in this repo