fix: shard sitemap.xml per locale

[myelinated-wackerow]

Summary

Part 2 of #18014, stacked on #18802 (base branch is fix/sitemap-static, not dev — review that first). Splits the sitemap into one shard per locale so it clears Google's per-file limits.

Why

Even after #18802 makes the sitemap build statically (complete, no truncation), it's still a single ~51 MB document with ~17.5k URLs and ~455k hreflang alternates — over Google's 50 MB / 50k-URL per-file limits, so crawlers reject it outright. Sharding is required for the sitemap to be accepted at all.

Changes

• app/sitemap.ts: generateSitemaps() emits one shard per locale (id = locale). The default sitemap({ id }) filters pages to that locale and keeps the full hreflang alternates block, so cross-language reciprocity is preserved. A memoized traversal lets all shards share a single content walk.
• app/robots.ts: lists every /sitemap/<locale>.xml shard. Next serves shards at /sitemap/<id>.xml and emits no index file, so robots.txt is the discovery mechanism; this replaces the old single /sitemap.xml reference.

Verification (full 25-locale build)

• All 25 shards prerender static (initialRevalidateSeconds: false in the prerender manifest — no ISR).
• Every shard closes with </urlset>, carries ~700 URLs, and includes the sentinel /wallets/find-wallet/ (the page that was invisible to crawlers in prod).
• Total across shards: 17,087 URLs. Largest shard ~2.01 MB (pt-br/zh-tw/en) — well under 50 MB, ~700 URLs each vs the 50k cap.

Post-merge follow-up (not code)

This removes the root /sitemap.xml. robots.txt handles crawler auto-discovery, but the sitemap registered in Google Search Console points at /sitemap.xml and should be updated to the shard URLs so GSC stops hitting a 404. Nothing in the codebase depends on the old path.

A CI regression guard (parse each shard: valid XML, closing </urlset>, URL-count floor, sentinel late-alphabet URL) will follow as a separate PR — this is the check that was missing when #17718 was closed on a single lucky fetch.

-- Opus 4.8

[github-actions]

🔎 First-pass review — ✅ Looks mergeable

Splits the sitemap into one shard per locale via generateSitemaps() (id = locale), keeps the full hreflang alternates block on every URL, and points robots.ts at the per-locale shard list. This is a sound, minimal way to get under Google's 50 MB / 50k-URL per-file caps, and the logic checks out: generateSitemaps() and the robots.ts shard list are both driven by LOCALES_CODES, so they cannot drift; the per-shard translatedLocales.includes(locale) filter partitions the same URL set the old single document emitted (total URL count is conserved, matching the ~17k figure). No convention issues — imports all used, no raw Intl.*, memoization is safe under force-static.

Note for reviewers: base branch is fix/sitemap-static (stacked on #18802) — review/merge that first. The GSC-registered /sitemap.xml will 404 after this ships; the described post-merge GSC update is required.

Analysis

• Lane: code (app/sitemap.ts, app/robots.ts).
• Checked: shard/robots consistency (both use LOCALES_CODES), URL-count conservation vs. the pre-shard doc, hreflang alternates preserved per URL, per-shard dedup (seenUrls), memoized traversal correctness under force-static, unused imports, chain-name / Intl wrapper conventions.
• Non-blocking: id is typed Promise<string> and awaited — runtime-safe either way; Next 16's generated route type check on the sitemap default export will confirm the signature (let CI settle). Empty shards for zero-page locales still emit valid <urlset></urlset>.
• Follow-up (author-noted, not code): CI regression guard parsing each shard is coming as a separate PR.

Generated by PR Reviewer (team) for #18812 · 88 AIC · ⌖ 24.2 AIC · ⊞ 5.2K · ◷