Accelerate Postgres Record List Reads

Why

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 record_json->>cursor_field, and count:"exact" spending about 1.1s scanning the full records table.

The Postgres schema already has stored cursor_value, primary_key_text, and retained-size projections, but the record-list path did not consistently use them and the write path did not populate cursor values for new rows.

What Changes

• Populate stored Postgres record sort positions from the manifest-declared cursor field on ingest.
• Backfill stored cursor values from registered connector manifests when a manifest is registered or refreshed.
• Order Postgres record-list pages by the stored sort-position columns instead of per-read JSON extraction.
• Satisfy unfiltered full-stream count:"exact" / count:"estimated" from a clean retained_size_stream projection, falling back to SQL count for filtered, time-ranged, resource-scoped, or dirty/missing projections.

Capabilities

Modified:

• reference-implementation-architecture

Impact

• Affects Postgres-backed query_records / REST record-list reads.
• Does not change public request or response shape.
• Requires an idempotent sort-position backfill on existing Postgres rows before the indexed read path can be treated as fully effective.