[Backport release-1.6] feat(kubevirt): expose migration configuration through platform values - #4339

europrinter (@yankawai) NOT LGTM because of one test case the backport dropped. Everything else matches #4254, and an existing 1.6 install renders exactly as before.

What I checked, against the current release-1.6 tip (8a6ce81):

• The diff differs from #4254's merge only where release-1.6 forces it. Platform values gain the kubevirt: key with its header, and the hop in iaas.yaml gets a reworded copy of the comment that sits over the disabledFeatureGates hop on main. The test script gains run_update_logged, and the Makefile header says "six". Main's rewording of a Makefile comment is left out, because release-1.6 doesn't have that paragraph. Nothing from the disabledFeatureGates work comes along, and the rest is line for line the same as main.
• make update from this branch, run on the release-1.6 CR, produces the committed CR byte for byte. A second run changes nothing. Without the new awk check, update_idempotency_test.sh goes red on the missing-key case, as the description says.
• With default values the rendered output is byte-identical between the base and this head. That covers the platform chart with values.yaml alone and with each of the isp-full, isp-full-generic and isp-hosted files, and the kubevirt chart with defaults and with a GPU-style set of forwarded values. I dropped templates/repository.yaml on both sides, since its lookup fails offline.
• With migrations set, the Package values and the rendered spec.configuration.migrations are the same as on main. I tried limits, false/0, {}, null, and kubevirt: null.
• kubevirt runs 9 tests (5 on the base), platform runs 101 in 24 suites (97 on the base), and the idempotency script passes. CI hasn't run any of this yet. The Pull Request, Pre-Commit Checks and API Review Gate runs on this head are waiting for workflow approval.

The blocker is a null-block test that was dropped together with the disabledFeatureGates tests around it. On main, #4254 added a notExists on spec.components.kubevirt.values.migrations to "renders the kubevirt Package without a nil-pointer error when the whole kubevirt block is set to null". That test came with the disabledFeatureGates work, so release-1.6 doesn't have it, and the assertion went with it. But on this branch (.Values.kubevirt).migrations is the only reader of .Values.kubevirt, so here the test belongs to this change.

Nothing on this branch catches a regression there. I changed iaas.yaml:17 to {{- with .Values.kubevirt.migrations -}}. The platform suite stays at 101/101, while helm template with kubevirt: null fails with nil pointer evaluating interface {}.migrations. The same edit on main turns that test red. The comment this PR adds over the hop says the walk is nil-safe for the reason the gpu block gives, and the gpu block says its null cases are pinned by tests. This one isn't.

The fix is to restore the case without its disabledFeatureGates assertion, next to the new migrations cases in bundles_gpu_kubevirt_wiring_test.yaml:

  - it: renders the kubevirt Package without a nil-pointer error when the whole kubevirt block is set to null
    set:
      bundles:
        system:
          enabled: true
          variant: isp-full
        iaas:
          enabled: true
        enabledPackages: []
      kubevirt: null
    documentSelector:
      path: metadata.name
      value: cozystack.kubevirt
    asserts:
      - equal:
          path: kind
          value: Package
      - notExists:
          path: spec.components.kubevirt.values.migrations

With it the suite passes at 102, and the flattened walk turns it red. The platform-defaults assertion dropped the same way doesn't need to come back. The default is {}, and "omits empty kubevirt.migrations" already fails if the hop forwards it. The wiring-test bullet in the commit message and in the description will need a small edit afterwards, since the description becomes the merge commit message on release-1.6.

Two things that don't block. The commit lists me as its author. It was cherry-picked from the merge commit of #4254, which I made, and cherry-pick keeps the author of that commit. You wrote the change and you sign it off, so git commit --amend --reset-author while you're amending puts your name on it.

The other is docs. The website's 1.6 page content/en/docs/v1.6/operations/configuration/platform-package.md is the table of platform values keys, and it has no kubevirt.* row. website#697 only touched next. The trigger map in docs/agents/contributing.md ties platform values.yaml to the next copy of that page, and says a docs change gets backported once it matters for a released version, which this one now does. A one-row follow-up there, linked from the description, fits better than the "no downstream repository" tick.

europrinter (@yankawai) LGTM. The test for a null kubevirt block is back, and it can go red.

It keeps the notExists check on migrations that #4254 has on main and drops only the disabledFeatureGates one, which has nothing to check on this line. I removed the parentheses at iaas.yaml:17, and only that test failed, with nil pointer evaluating interface {}.migrations. With the parentheses back the platform suite passes, 102 tests.

The rebuilt commit sits on the same release-1.6 base, and the author field is now yours.