Refactor toplevel site.yml and fatimage.yml playbooks

[sjpb]

Refactoring to resolve a number of problems and inconsistencies between the site.yml and fatimage.yml playbooks. The major ones are:

• iam.yml and slurm.yml playbooks are now run directly from fatimage.yml to make making changes to these less error-prone.
• compute-init export task moved to a new final.yml playbook so it will run after any role which generates configuration which it might require. install.yml task for this role also been moved here for consistency - note this changes commands used for development.
• gateway role which installs ansible-init playbook moved adjacent to ansible-init install
• Ordering in site.yml and fatimage.yml is now consistent in that repos are disabled after any post-hook in both cases.
• Unrunnable build VM cleanup code removed from site.yml.
• Correctly define dependency of roles injecting ansible-init playbooks (compute_init, gateway) in groups hierarchy.

See https://wiki.stackhpc.com/doc/slurm-top-level-playbooks-HW9BlHDMpm for more detail

TODO:

• [ ]: Fix whatever change in mysql means the slurm db isn't getting persisted - probably writing the config during build or something with the appliances_state_directory not set or something.

[sjpb]

Build for the above image bump: https://github.com/stackhpc/ansible-slurm-appliance/actions/runs/14660430262/job/41143461025

[sjpb]

Interactive testing @ 3120e49:

✔️ ondemand shell
✔️ ondemand remote desktop
✔️ ondemand files:
✔️ ondemand jupyter
✔️ ondemand monitoring

[wtripp180901]

Looks good, a few suggestions for keeping the pattern of how appliances_mode is used consistent. As you said it may also worth documenting how appliances_mode is intended to be used vs builder groups

[sjpb]

Testing image build - should rebuild once fix/zenith merged to main anyway: https://github.com/stackhpc/ansible-slurm-appliance/actions/runs/14778751137

[sjpb]

Image build: https://github.com/stackhpc/ansible-slurm-appliance/actions/runs/14878191724