Accelerate Postgres Record List Reads

Large Postgres-backed record streams were not "blazing fast" on the public read surface. Live evidence on pdpp.vivid.fish showed a Slack messages page spending about 1.3s sorting recordjson->>cursorfield, and count:"exact" spending about 1.1s scanning the full records table.

The reference Postgres store has grown large enough that JSON-derived ordering and whole-table exact counts are owner-visible latency. Live EXPLAIN ANALYZE showed:

The Postgres-backed reference record-list path SHALL preserve the public record-list contract while avoiding per-request full-stream work when an equivalent maintained read model exists.