fix(k0s): select RHEL 9 NVIDIA DKMS stream for air-gap installs

[kbhos-splunk]

Description

Fix RHEL 9 air-gap NVIDIA closure staging when the CUDA repository's default
open-dkms module stream hides the established
kmod-nvidia-latest-dkms package from a plain dnf download operation.

The air-gap installer now checks whether the RPM is already visible and, only
for RHEL 9, resets a conflicting/default NVIDIA stream and enables
nvidia-driver:latest-dkms before resolving the existing package closure.
RHEL 10 continues using ordinary RPMs, Ubuntu 24.04 continues using its APT
container path, and the non-air-gap installer is unchanged.

Related Issues

• No linked issue

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactoring (no functional changes)
• Performance improvement
• Test improvement
• CI/CD improvement
• Chore (dependency updates, etc.)

Changes Made

• Add a RHEL 9-only helper that selects nvidia-driver:latest-dkms when
  kmod-nvidia-latest-dkms is hidden by modular filtering.
• Preserve an already-compatible module selection and verify that the package
  becomes visible after changing streams.
• Add mocked regression coverage for hidden and already-visible RHEL 9 streams,
  RHEL 10, Ubuntu 24.04, and non-air-gap isolation.
• Document the automatic stream selection and update the manual closure recipe.

Testing Performed

• Unit tests pass (bash tools/cluster_setup/test_k0s_cluster_with_stack.sh — 246 passed, 0 failed)
• Linting passes (shellcheck is not installed in the development environment)
• Integration tests pass (not run)
• E2E tests pass (not run)
• Manual testing performed (verified live NVIDIA RHEL 9/RHEL 10 repository and module metadata)

Test Environment

• Kubernetes Version: Not applicable to unit tests
• Cloud Provider: AWS EC2
• Deployment Method: k0s air-gap installer

Test Steps

1. Run bash -n against the modified installer and test scripts.
2. Run bash tools/cluster_setup/test_k0s_cluster_with_stack.sh.
3. Run bash tools/cluster_setup/test_installer_dry_run.sh.
4. Verify the RHEL 9 module metadata defaults to open-dkms while
   latest-dkms remains available as a separate stream.
5. Verify the RHEL 10 repository publishes kmod-nvidia-latest-dkms as an
   ordinary RPM and therefore bypasses the RHEL 9-only helper.

Documentation

• Updated inline code comments
• Updated README.md
• Updated API documentation
• Updated deployment guides
• Updated CHANGELOG.md
• No documentation needed

Checklist

• My code follows the project's style guidelines
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings in the executed test suites
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published (none required)
• I have updated the Helm chart version (not applicable)
• I have updated CRD schemas (not applicable)

Breaking Changes

Impact: None. The new behavior is limited to RHEL 9 air-gap NVIDIA closure
resolution when the requested driver RPM is not already visible.

Migration Path: None required.

Screenshots/Recordings

Not applicable.

Additional Notes

The original failure occurred before cluster installation while building the
NVIDIA closure. An end-to-end air-gap cluster installation should be performed
before merging to confirm the repository closure resolves and installs on the
target GPU nodes.

Reviewer Notes

Please pay special attention to:

• The RHEL 9-only guard around DNF module changes.
• Preservation of the RHEL 10, Ubuntu 24.04, and non-air-gap code paths.
• Whether changing the build host's selected NVIDIA module stream should be
  restored after closure creation.

---

Commit Message Convention: This PR follows Conventional Commits