Make it easier to run this on the cluster

[gaurav]

Note

Built on PR #9; open until it has run on ht1 for real.

slurm/ now holds one sbatch script per step (01-download … 05-validate), a config.sh of shared settings, and a submit.sh:

./slurm/submit.sh load              # one step; read its log, then decide
./slurm/submit.sh all               # chained with --dependency=afterok
./slurm/submit.sh --dry-run all     # print the sbatch commands, submit nothing

Steps stay separate jobs because their resource shapes are nothing alike — one allocation sized for all three would hold the export's big node for the load's several hours. --dependency=afterok cancels the rest of a chain when a step fails, which is the automated form of reading each log first, but it only knows about exit status, so one-step-at-a-time is still the careful mode and is fully supported.

What is unverified: everything against a real sbatch. It was exercised through --dry-run and a stubbed uv; no job has been submitted. The dependency chain in particular is verified only as the command lines it would produce. Worth watching the first ./slurm/submit.sh all with squeue. Closing this is the job of that run.

---

We already have a slurm/README.md that has all the commands and information needed to run this on the cluster, but we might be able to simplify this further with an sbatch script or a runner script or something.