refactor: extract landmark-roles util (preserving deliberate per-rule region exclusion)

[johanrd]

Three a11y rules share a concept of "landmark ARIA role." Today they each maintain their own hand-list. The counts look like drift:

Rule	Count	Missing
template-no-redundant-role	7	region
template-no-nested-landmark	7	region
template-no-duplicate-landmark-elements	8	—

But the two 7-role lists aren't drift — the region exclusion is deliberate, documented, and was landed in #2694 after a "what's MDN say?" review. The reasoning:

<section> only gets the region landmark role when it has an accessible name (aria-label / aria-labelledby / title). Without one, <section> has role generic. A static linter cannot verify at lint time whether aria-label resolves to a non-empty name at runtime.

Source: WAI-ARIA APG HTML5 landmark examples.

The third rule (template-no-duplicate-landmark-elements) DOES include region because that rule inspects aria-label / aria-labelledby before classifying a node as a landmark — it can safely handle the full 8-role set.

What this PR does

Adds lib/utils/landmark-roles.js that makes the distinction explicit by exporting two sets:

• ALL_LANDMARK_ROLES — the 8 WAI-ARIA 1.2 §5.3.4 landmark roles, derived from aria-query (non-abstract roles whose superClass chain includes 'landmark', minus DPub-ARIA doc-* roles).
• LANDMARK_ROLES — the 7-role subset excluding region. The safe default for static-linting rules that can't verify accessible names.

Both sets are derived from aria-query's role metadata so future WAI-ARIA additions flow through on aria-query upgrade.

Rule migrations:

• template-no-duplicate-landmark-elements → ALL_LANDMARK_ROLES (verifies accessible names).
• template-no-nested-landmark → LANDMARK_ROLES (preserves Post-merge-review: Fix template-no-nested-landmark: drop port-only section/region #2694 behavior).
• template-no-redundant-role → LANDMARK_ROLES (preserves existing landmark-only behavior).

Also removes the region: ['section'] entry from template-no-redundant-role's ROLE_TO_ELEMENTS mapping, since #2694's reasoning applies to that rule too: <section role="region"> shouldn't be flagged as redundant when we can't verify the section has an accessible name.

Behavior vs. master

• template-no-duplicate-landmark-elements: no change — same 8 roles.
• template-no-nested-landmark: no change — same 7 roles.
• template-no-redundant-role: <section role="region"> is no longer flagged as redundant (default mode). This aligns the rule with Post-merge-review: Fix template-no-nested-landmark: drop port-only section/region #2694's reasoning.

Spec references

• WAI-ARIA 1.2 §5.3.4 Landmark Roles
• WAI-ARIA APG HTML5 landmark examples — §section-without-name-is-generic basis for region exclusion
• #2694 — precedent for region exclusion

[NullVoxPopuli]

why does this need a function now where it didn't before?

[johanrd]

The set is now computed by being derived by aria-query's role map rather than being hand-maintained. This aligns with other external-lib derivations and picks up any new landmark roles aria-query adds automatically.

[NullVoxPopuli]

wouldn't importing it also do that? I don't understand

[johanrd]

aria-query doesn't expose a landmark-roles list as a top-level export. The landmark role only appears as a superClass entry inside the per-role definitions in roles. So to get "every non-abstract role that descends from landmark" you have to walk roles and filter on superClass, which is what buildLandmarkRoleSet() does. If aria-query ever adds a direct export we can swap to that.

[NullVoxPopuli]

gotchya, thanks!