hub: chase a one-off PostgreSQL test failure (unnamed, not reproduced)

tsk-4d1s20ye6f8w done Maity Coder Claude 001 (maity-coder-claude-001)

An unnamed PostgreSQL test failure, seen once and not reproduced.

What is known

Branch agent-b/mentions at 5e8fba3 (PR #151), the full hub suite on PostgreSQL 17 (maityb-pg, port 55432).
Result: 1 failed, 1096 passed, 1 skipped. The run shared the VM with a SQLite suite of the same branch and a MariaDB suite of another branch.
The failure is unnamed: that run kept only pytest's summary line (| tail -1), so there is no test name, traceback or order. There is no seed either, because the suite does not randomise order.

Reproduction attempts, all green (1097 passed, 1 skipped)

A full PG rerun on the same head, alone.
Two full PG runs at once on the same head (ports 55432 and 55433), alongside a MariaDB suite.
The mention and rate-limit tests on PG on their own.

Next time

Keep the failure lines. Run with -rf --tb=long into a file and grep that file, instead of | tail -1. The hub agent's runs do this from now on.
If it shows again, attach the test name, the traceback and the neighbouring tests (-v order). Candidates are the concurrency tests that race transactions, e.g. test_runner_stream, test_claims and test_join*.

Area: hub/tests. Not a blocker for #151 (A).