[P0] Align public pagination regression assertions with interactive-lift

Product outcome

Restore the public pagination regression contract so the scheduled/full Django gate reflects the accepted shared pagination markup on current main. This is a test-contract correction only: the rendered paginator already intentionally carries the accepted interactive-lift design-system token.

Evidence and ownership boundary

• On current origin/main face8e4808d65afbf0374d1ced7a88079950d663, content/tests/test_public_pagination.py:423-424 still counts the obsolete exact strings:
  • class="filter-pill pagination-number"
  • class="filter-pill pagination-step"
• templates/public/_pagination.html:46,52,62 emits the accepted classes with interactive-lift appended.
• The token was introduced by the accepted a5d08f1 (“Unify clickable card and control interactions”), not by a paginator behavior change.
• Scheduled run 33513872199 and the integrated v3 Django evidence reproduce the failure; the first failing assertion is line 423.
• The collection-hub/archive class correction remains in #261, whose groomed scope explicitly limits its focused test changes to test_archive_row.py and test_collection_hub.py. #280 is CI evidence orchestration and explicitly excludes product/test pagination changes. This issue therefore owns only the separate test_public_pagination.py stale expectations.
• Shared paginator behavior and its accepted design-system contract are defined by closed #178.

Scope

Change only content/tests/test_public_pagination.py:

1. Update the numeric-control count to match class="filter-pill pagination-number interactive-lift" and retain the expected count of 7.
2. Update the direction-control count to match class="filter-pill pagination-step interactive-lift" and retain the expected count of 2.
3. Keep every other pagination assertion unchanged, including control counts, labels, aria-current, arrows, page-window/ellipsis behavior, URLs, canonical/SEO/cache semantics, boundary behavior, empty/single-page behavior, and source/design-system guards.

Acceptance criteria

• The candidate changes no production template, CSS, Python application code, fixture data, browser harness, generated projection, or #261/#280-owned path.
• The two rendered-class assertions recognize the accepted interactive-lift token and still prove exactly 7 numeric controls and 2 direction controls on the 12-page middle state.
• No assertion is weakened by removing the control class, reducing the expected count, using a broad substring that can match unrelated markup, or changing the paginator to satisfy the test.
• content.tests.test_public_pagination passes on the frozen candidate, including all existing query grammar, status, canonical, cache, accessibility, URL-window, boundary, and source-contract cases.
• The engineer runs the focused module plus make format-check, make lint, and the applicable versioned verification checks; an independent tester recomputes the plan and verifies the same frozen candidate. Final full-Django/full-regression evidence must be green after the separately owned #253/#261 and other release blockers are repaired; this issue never waives a red gate.
• Screenshots are not_applicable: only a regression test file changes and no rendered product input changes.

Non-goals

• No paginator implementation, template, CSS, accessibility, URL, SEO, copy, or visual change.
• No archive-row/collection-hub class work, normalization/type work, formatter repair, or structured-evidence workflow work from #261/#280.
• No test deletion, skip, xfail, quarantine, timeout/retry change, assertion weakening, or broad test refactor.
• No commit, push, merge, deployment, or workflow rerun as implementation evidence before engineer/tester/PM gates.

Dependencies and handoff

No implementation dependency. This is a small independent testing/frontend lane and may be implemented without waiting for #261 because the changed file does not overlap #261's allowed paths. Final release verification remains subject to the common green-base/order recorded in #253/#261 and the release coordinator.

The engineer works uncommitted in an isolated worktree, posts the versioned handoff, and freezes it. A separate tester performs the focused and graph-selected checks. PM accepts only from a passing tester envelope; the orchestrator then owns commit/merge/push under _docs/PROCESS.md.

Start in content/tests/test_public_pagination.py at assertions around lines 423-424, then compare them with the classes emitted at templates/public/_pagination.html lines 46, 52, and 62. Run content.tests.test_public_pagination first, followed by make format-check, make lint, and applicable versioned checks; done means only the two assertions recognize interactive-lift while all existing checks pass.