i18n/seo: keep locale out of paths while defining cache and indexing behavior

[tiankaima]

Product decision

Keep locale out of semantic paths and query strings. Existing unprefixed URLs remain canonical. Language is selected from the saved preference/cookie and normalized Accept-Language for anonymous requests.

This has an unavoidable SEO trade-off: the same URL cannot provide two independently crawlable canonical language pages or truthful reciprocal hreflang alternates. For now, Chinese is the default canonical/indexable representation; English is a complete adaptive user experience, not a separately guaranteed search index.

If independently indexed English becomes a requirement later, evaluate locale-specific hosts (for example en.example) as a separate product/operations decision. Do not add /zh-cn, /en-us, or locale query parameters.

Cache policy

• Authenticated/personalized HTML remains private/no-store.
• Anonymous caching, when [perf] Introduce an authenticated gateway for anonymous SSR edge caching #507 is implemented, normalizes locale before the cached entrypoint and includes the normalized locale in the internal representation key.
• Never key shared HTML on raw Cookie or raw Accept-Language.
• Emit correct html lang, Content-Language, localized title/description/social metadata, and a single self-canonical URL.
• Use Vary consistently on negotiable APIs; do not emit misleading hreflang links to identical URLs.

Acceptance criteria

• No locale path/query route is introduced.
• Cookie, header, and explicit user preference produce aligned SSR HTML, metadata, and hydration.
• Anonymous cache tests prove zh/en isolation and authenticated bypass.
• The sitemap contains each semantic URL once; canonical URLs are stable.
• Search documentation clearly states Chinese-indexing and English-adaptive limitations.
• The language selector remains accessible and persists preference without creating duplicate canonical pages.