Wave-4 fix-batch decomposition: seven pg-fix tasks + review-wave-4-fixes gate from the general review (002) — forwarder reconnect retry (Finding 1, with the failed-connect test seam), tx pg_notify wakes retiring F-1's engine arm (Finding 2), max_size/DSN-options open path (Finding 3), stale-UNLISTEN generation drop (Finding 4), scheduler quarantine/retry (Finding 5), decode dedupe + cleanups, doc alignment; wave 5 gates on the two HIGH fixes landing

tasks/pg-fix-dedupe-cleanup.md

@@ -0,0 +1,83 @@
+---
+id: pg-fix-dedupe-cleanup
+name: Decode dedupe + small cleanups in queue/tx/notify (review 002 minor notes)
+status: pending
+depends_on: [pg-fix-tx-wake]
+scope: narrow
+risk: low
+impact: component
+level: implementation
+tags: [wave-4-fixes, postgres-engine, cleanup]
+---
+
+## Description
+
+Batch the wave-4 general review's minor notes that share files and are
+too small for standalone tasks
+(`docs/reviews/002-wave-4-general-review.md` §Minor notes + §Finding
+6's queue bullet):
+
+- **`job_from_row` is byte-for-byte duplicated** (`queue.rs:411` and
+  `tx.rs:626`; only `pub(crate)` differs), and `get_job_tx` inlines
+  the 18-column dead/live select column lists (`tx.rs:420-428,
+  438-442`) duplicating `queue.rs`'s `live_columns()`/`dead_columns()`.
+  Both files claim "one decode owner" in their doc comments; both are
+  wrong. The `Job` shape is contract-pinned, so two owners is exactly
+  where a shape edit goes wrong later. **Dedupe into one module** —
+  `queue.rs` is the natural home; `tx.rs` imports it (as `tx.rs`
+  already imports `stream_events_from_rows` from `stream.rs`). The
+  doc comments must again tell the truth (one decode owner).
+- **`queue.rs:71` doc fix (Finding 6)**: `sweep_expired` "returns the
+  moved count" — it returns moved + retention-deleted (`queue.rs:720`).
+  The SQLite twin returns the same sum (substrate `queue_ops.rs:232-243`),
+  so the engines agree; only the doc is wrong. ("Single-statement
+  atomic" is also inaccurate — it's the `in_tx` multi-statement
+  frame.) Correct both claims.
+- **`notify.rs:106-110`**: the closed-store error is built inline
+  (`Error::database(std::io::Error::other(...))`) instead of the
+  shared `database_error` helper — one duplicate of the one message.
+  Route it through the helper.
+- **`bridge_capacity()`** (`notify.rs:137-139`) is a function
+  returning a constant — a `const` says it plainly. Convert (keep the
+  doc comment's rationale on the const).
+
+No behavior changes anywhere in this task — it is dedupe + doc
+truth-telling. All existing tests must pass unchanged; where a test
+asserts on the moved-count semantics, it already pins the sum (the
+doc was the only liar).
+
+**Depends on `pg-fix-tx-wake`**: that task edits `tx.rs`'s
+publish/enqueue paths; this task restructures `tx.rs`'s decode side —
+landing after avoids touching the same file regions blind.
+
+## Acceptance Criteria
+
+- [ ] `job_from_row` has one owner (`queue.rs`); `tx.rs` imports it;
+      both files' "one decode owner" doc claims are true
+- [ ] `get_job_tx`'s dead/live column lists reuse `queue.rs`'s
+      `live_columns()`/`dead_columns()` (no inlined duplicates)
+- [ ] `sweep_expired`'s doc states the moved + retention-deleted sum
+      and the `in_tx` frame (matching the SQLite twin's semantics)
+- [ ] `notify.rs`'s closed-store error routes through
+      `database_error`; `bridge_capacity` is a `const`
+- [ ] Zero behavior change: the full pg lib suite green against the
+      harness server, unchanged assertions
+- [ ] `cargo clippy --all-targets -- -D warnings`, fmt clean; gates
+      green server-less
+
+## References
+
+- docs/reviews/002-wave-4-general-review.md §Minor notes + §Finding 6
+  (the queue bullet)
+- docs/architecture/core-contract.md (the pinned `Job` shape — why one
+  decode owner matters)
+- tasks/pg-engine-queues.md, tasks/pg-engine-seam-tx.md (the tasks of
+  record for the touched files)
+
+## Notes
+
+> To be filled by implementation agent
+
+## Summary
+
+> To be filled on completion