Add role to install NVIDIA DOCA on top of an existing "fat" image

[sjpb]

• Adds new doca role to install NVIDIA DOCA:

  • This is only run during image builds where inventory_groups include doca.
  • It may be run on an existing "fat image" e.g. as part of an "extra" build.
  • It rebuilds the DOCA kernel packages to match the installed kernel.

  This role should be preferred over the ofed role which may be deprecated at a later date.
  NB: This uses doca packages from upstream repos, not StackHPC's ark.

• Adds a workflow to test the DOCA build during CI on the current RL8/RL9 fat images.
  Unless run manually, the built image is deleted on completion.
  NB: the resulting DOCA image is not tested by CI.

• Simplifies the configuration for packer builds:

  • Only a single build is now defined. Control of behaviour is via variables, rather than by selecting a build.
  • All variables are now simple (instead of some being maps keyed by the OS version).
  • The inventory groups to add the build VM is now specified via inventory_groups, taking a comma-separated list (instead groups which took a map).

• All image build workflows have been adjusted to use the new packer configuration

The following manually-triggered checks related to packer configuration changes have been completed:

• Check build works with empty groups
• Check fatimage workflow works
• Check nightly workflow works

Ticket: PLATFORM-537

[sjpb]

Fat image build: https://github.com/stackhpc/ansible-slurm-appliance/actions/runs/12276360419

[sjpb]

Fat image build: https://github.com/stackhpc/ansible-slurm-appliance/actions/runs/12277251098. NB: doca workflows fail here b/c previous fat image has signature_verified property.

[sjpb]

Note the above didn't actually build DOCA for some reason even though the workflow ran with the right groups

[sjpb]

Tested nightly build here: https://github.com/stackhpc/ansible-slurm-appliance/actions/runs/12280868395
At least started running ansible and checked it had correct groups, cancelled it before it could produce an image.

[bertiethorpe]

Packer and Workflows LGTM

[bertiethorpe]

Docs LGTM