fix(questions): correct inverted minus scoring for zh-cn and vi in the published 1.6.x source

[csliubo]

Fixes the zh-cn and vi scoring bug documented in #31 — retargeted to the actually-published source after tracing the release pipeline.

What & where

The live @bigfive-org/questions@1.6.4 is built from old-packages/questions/ (auto-published to npm with --tag beta by questions-publish.yaml on push), not from packages/questions/. This PR therefore patches:

1. old-packages/questions/data/zh-cn/choices.js — the minus array had its labels in reversed order while keeping scores 5..1, so on all 55 reverse-keyed items 非常不准确 (Very Inaccurate) recorded 1 instead of 5. Roughly half of every zh-cn result is scored backwards; the all-minus facets (A2 Morality, A4 Cooperation, A5 Modesty, C6 Cautiousness) come out fully inverted. Restored the invariant used by en and the other 39 correct locales: minus lists the same labels in the same display order as plus, scores 5..1. Also replaced the score-2 label 不太不准确 (an ungrammatical double negative) with 比较不准确 to match Moderately Inaccurate.
2. old-packages/questions/data/vi/choices.js — identical defect, same structural fix.
3. old-packages/questions/package.json — 1.6.4 → 1.6.5. Two reasons: npm won't accept a same-version republish, and the workflow's paths filter old-packages/questions/* doesn't match nested files (single * doesn't cross /), so a data-only change would not trigger the publish — the package.json change does. If you'd rather control versioning yourself, say the word and I'll drop this hunk.

Verified end-to-end against a local build of the site

Linked web/ to the patched package via a pnpm override and fetched /zh-cn/test from a local dev server. Served payload after the fix:

plus:  非常不准确→1 … 非常准确→5   (unchanged, correct)
minus: 非常不准确→5 … 非常准确→1   (fixed; live site currently serves 非常准确→5 first)

Follow-up for maintainers

• web/package.json pins @bigfive-org/questions: 1.6.4 — it needs a bump to 1.6.5 after CI publishes the new version. Left out of this PR deliberately: bundling it would race the Vercel build against the npm publish in the same merge.
• Historical zh-cn / vi results stored since the 1.6.x rollout cannot be distinguished from clean ones without raw answers; a note to affected users may be worth considering.
• Full forensics (npm diff, live payload capture, 41-locale structural scan): [Bug] zh-cn and vi: all 55 reverse-keyed items are scored inverted on the live site (@bigfive-org/questions 1.6.4) #31.

🤖 Generated with Claude Code — analysis and patch produced with AI assistance, disclosed in the commit trailers.

https://claude.ai/code/session_01C5EiE3nDRxWdHLt3r3xgpx

[vercel]

@headlessQ is attempting to deploy a commit to the Big Five Team on Vercel.

A member of the Team first needs to authorize it.

[csliubo]

Closing this for now while I double-check the fix against the actual release pipeline — will reopen (or resubmit) once verified. The full bug forensics remain in #31.