[Test] In test_patching_cluster, reduce the risk of ICEs and facilitate troubleshooting

[gmarciani]

Description of changes

In test_patching_cluster:

1. reduce the risk of ICEs by using c5.xlarge for headnode and login nodes rather than GPU instances.
2. increase the patching timeout for the head node to 3600s (from 1800s) to avoid the risk of timeouts when small instance types are used (such as c5.xlarge).
3. print the full ImageBuilder log when the patching fails to facilitate troubleshooting.
4. support new patching flavor minimal-capped and full-capped, where the patching is capped to the maximum kernel version supported by Lustre.
5. in both develop and released test config, execute full-capped on both Ubuntu and RHEL/Rocky, and full (uncapped) on AL2023, where capping is not required.

Notes

1. skip-bad-url-suffix-check: safe to skip because the check would otherwise fail for the expected FSx Lsytre repo URL

Tests

ONGOING

test-suites:
  patching:
    test_patching.py::test_patching_cluster:
      dimensions:
        - regions: [{{ g4dn_8xlarge_CAPACITY_RESERVATION_2_INSTANCES_2_HOURS_NOPG_ubuntu2404 }}]
          instances: ["g4dn.8xlarge"]
          oss: ["ubuntu2404"]
          flags: ["patching:minimal-capped"]
          schedulers: ["slurm"]
        - regions: [{{ g4dn_8xlarge_CAPACITY_RESERVATION_2_INSTANCES_2_HOURS_NOPG_rhel9 }}]
          instances: ["g4dn.8xlarge"]
          oss: ["rhel9"]
          flags: ["patching:minimal-capped"]
          schedulers: ["slurm"]

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.