questionFeed: de-hardcode the fixed question ladder (issue #728)

[WilfordGrimley]

Summary

Issue #728 — the stage machine hardcoded a fixed question ladder (level1 → level2 → level3), forcing every confirm_suggestion item through a redundant funnel. The de-laddered feed now has a single candidate question; answers resolve or advance by what they mean, not by a hardcoded sequence.

1. QuestionFeed.tsx — remove the fixed ladder. CandidateStage and initialStage() are gone; the only remaining "stage" is the data-driven post-selection attribute confirmation (level3Active, entered only when a selected candidate leaves an exclusion group open). Every new item starts outside it.

   • The suggested printing (when present) is judged exactly once, in its own slot on page one (question-feed-suggestion-*), and is never re-offered as a grid tile — the old Level 2 re-presented the same candidate "highlighted", i.e. asked about twice.
   • The rest of the candidates render in the grid on the same page (question-feed-candidate-question), so there is no level1→level2 funnel for non-yes answers: "No" collapses the suggestion slot into a "you said not this one" context line with the grid still carrying the question; "Not sure" records an abstention and advances.
   • "Art matches, not an official printing" is now a first-class, page-one fallback in the bottom row (question-feed-custom-art) instead of a Level-2-only escape hatch.
   • The singleton case is preserved: rejecting the only candidate immediately casts the terminal no-match vote.

2. QuestionFeed.test.tsx — ported stage-machine tests. The Level-1 "Not sure" test now asserts the de-laddered advance (abstention + next fetch, no level-2 re-ask); selectors renamed to question-feed-suggestion-*.

3. QuestionFeed.spec.ts — spec updated to the new flow. The confirm_suggestion specs assert the single-page shape (suggestion slot + grid, suggested candidate never a tile). Two stale specs that still encoded the old ladder were fixed: the singleton-NO spec (question-feed-level1-no → question-feed-suggestion-no) and the 390px mobile overlap spec (question-feed-level1-card-panel → question-feed-card-panel, control list now the suggestion slot + bottom row).

4. Downstream specs that keyed off the removed level-1 testids — PDFWaitExperience.spec.ts, QuestionFeedResponsive.spec.ts, CardbackPdfWaitFidelity.spec.ts — updated to the renamed surface (question-feed-candidate-question, question-feed-suggestion-*). The question feed is embedded in the PDF-wait game and asserted by these specs, so they broke the moment the ladder was removed.

Test plan

• Tests updated/added: unit + Playwright specs ported to the de-laddered flow; two stale ladder specs rewritten to cover the new shape; three downstream specs fixed to the renamed selectors
• Unit: npx jest src/features/questionFeed/QuestionFeed.test.tsx — 23 passed
• Playwright: npx playwright test tests/QuestionFeed.spec.ts — 28 passed
• Playwright: npx playwright test tests/PDFWaitExperience.spec.ts — 3 passed
• Playwright: npx playwright test tests/QuestionFeedResponsive.spec.ts tests/CardbackPdfWaitFidelity.spec.ts — 30 passed
• Playwright: npx playwright test tests/ModerationTab.spec.ts tests/NoMatchReasonStrip.spec.ts tests/WhatsThatPWA.spec.ts tests/WhatsThatWordsAnimation.spec.ts — 20 passed
• Lint: prettier --check (pre-commit-pinned 2.7.1) passes on all 6 touched files; eslint 0 errors (5 pre-existing no-img-element warnings, identical count on master); local pre-commit gate passed on commit
• Build: npm run build (Next.js static export) succeeds
• Manual: on /whatsthat with a live backend, a confirm_suggestion item shows the suggested match in its own slot with the remaining candidates on the same page; "No" collapses the slot (never re-presented); the bottom row resolves the whole question from page one
• CI — not watched by this session per dispatch convention

Closes #728