#533 feat: S3/GCS extension namespace-scoped sync support

Problem

The core giga-swamp infrastructure is complete (Phases 1-4, 6 shipped in swamp). The DatastoreSyncOptions.namespace field flows through the sync coordinator and acquireModelLocks, the namespacedSync capability flag exists, and the sync service interface has exportCatalog/pullForeignCatalogs/fetchForeignContent methods. But no extension implements them yet — push/pull still syncs everything regardless of namespace.

Until the S3 extension supports namespace-scoped sync, multi-repo shared datastores dont work end-to-end.

What needs to change

1. Per-namespace index partitioning

When options.namespace is set, scope the index walk to the {namespace}/ prefix. Use {namespace}/.datastore-index.json instead of the global one.

• pullChanged: fetch {namespace}/.datastore-index.json, download only keys under {namespace}/
• pushChanged: upload only files under {namespace}/, update {namespace}/.datastore-index.json
• Solo mode (no namespace): use global index as today

2. exportCatalog(namespace)

After pushChanged, write .catalog-export.json to {namespace}/ in the bucket. Flat JSON array of CatalogExportRow.

3. pullForeignCatalogs(namespaces)

Fetch {namespace}/.catalog-export.json from each namespace. Return CatalogExportEntry[]. Skip missing exports silently.

4. fetchForeignContent(namespace, relPath)

Download single file from foreign namespace. Return bytes or null.

5. Advertise namespacedSync capability

capabilities() returns namespacedSync: true.

Critical constraints

• Do not delete files outside the namespace prefix during push
• markDirty contract unchanged
• Backward compatible with solo mode global index
• Solo mode regression: existing workflows must work identically

Verification

All UAT scenarios from swamp-club#525 must pass with the extension update.

Design Context

Core: swamp-club#525 (PR #1493). Design doc: resources/giga-swamp.md