Data: cursor pagination for stable iteration over large collections

[jwicks31]

What

Adds opt-in cursor pagination alongside the existing limit/offset. A list response now carries a nextCursor; pass it back as ?after=<cursor> to continue in stable (sort, id) order — no skips or repeats when rows are inserted mid-iteration, the way offset drifts. Built to pair with the declared indexes so it stays fast at any size.

How

• The cursor encodes { sortField, dir, value, id } (base64url). The next page adds a row-value predicate (sortField, id) > (value, id) (flipped to < for DESC), reusing the dialect's typed JSON comparison so it's correct on SQLite (json_extract) and Postgres ((data->'f') vs to_jsonb(?::type)). id is the unique tiebreaker and is appended to ORDER BY for a deterministic total order.
• A malformed cursor — or one whose encoded sort doesn't match the request's sort — is a clean 400.
• list() now returns { documents, nextCursor }; offset paging is unchanged when no cursor is supplied. SDK: page({ after }), buildQuery emits after, and Query.after / Page.nextCursor are typed.

Verification

• SQLite 64/64 including a test that walks a collection page-by-page via the cursor while inserting a new row mid-iteration — every original row is seen exactly once with no duplicates, and the late insert sorts into place; plus malformed-cursor and sort-mismatch 400s.
• Postgres 46/46 (Postgres 16) — the test sorts by a JSON field, exercising the to_jsonb comparison path.
• Docs updated: README (data section) and /docs (REST table + queries).

This completes both items from your steer — per-collection schemas (#77) and now cursor pagination.

https://claude.ai/code/session_018efxvWw3MRjdtvE5xgBqya

---

Generated by Claude Code