Support additional nodegroups

[sjpb]

• Adds opentofu variable additional_nodegroups to support defining non-Slurm nodes in the cluster. E.g.:

  additional_nodegroups = {
      squid = {
          nodes = ["squid-0"]
          flavor = var.other_node_flavor
      }
  }

  Nodes are automatically added to an inventory group of the same name as the node group.
  Slurm-controlled rebuild and compute-init is not supported, as these nodes will not be running slurmd.
  Security groups are those from opentofu variable nonlogin_security_groups by default, but may be overriden.

• Also adds compute nodes into an inventory group of the same name as the note group, in addition to the existing ${cluster_name}_${group_name} inventory group required for stackhpc.openhpc role partition configuration. This simplifies multi-environment configuration.

[bertiethorpe]

LGTM, once CI passing. I agree that login and compute should be suffixed with "_nodegroups" but we can live with it. It would be good to test this in CI at some point too

[sjpb]

Should have said, this has been tested locally using the CI environment, so it does actually work! Bit loath to add it to CI as we already struggle with resources.

[sjpb]

First attempt above - RL9 (only) failed with:

# Configure cluster at latest release
TASK [Run sinfo] ***************************************************************
...
FAILED - RETRYING: [slurmci-RL9-2607-login-0]: Run sinfo (1 retries left).
fatal: [slurmci-RL9-2607-login-0]: FAILED! => {
    "attempts": 200,
    "changed": false,
    "cmd": "sinfo --noheader --format=\"%N %P %a %l %D %t\" | sort",
    "delta": "0:00:00.013515",
    "end": "2025-06-24 16:50:16.721362",
    "rc": 0,
    "start": "2025-06-24 16:50:16.707847"
}

STDOUT:

 extra up 60-00:00:00 0 n/a
slurmci-RL9-2607-compute-[0-1] standard* up 60-00:00:00 2 drain