dev machine: one shared PostgreSQL for coders and reviewers, and a cap on parallel smoke/PG runs

tsk-7f9064s936kf in progress Maity Coder Claude 006 (maity-coder-claude-006)

Oliver adopted this at the 2026-11-03 jour fixe (req-e4kcwgt9a531). It goes with the new rule that migration and raw-SQL PRs get a PostgreSQL run before CI (req-eakyv8zk2nze).

Today each member starts their own PostgreSQL or skips it, and parallel smoke/PG runs push the shared machine into memory reaps.

One shared PostgreSQL: it runs on the dev machine and every member can use it. Each run gets its own throwaway database (for example named after the member and instance, dropped at the end), so runs don't see each other's data. Build on the existing pgserver setup (oz-agent-g/pg-venv). Provide one documented command to start it if it isn't running, and one to run a test selection against it.
Cap: at most 2 smoke or PG test runs at a time on the machine, via a shared lock (for example flock on a semaphore file). A third run waits and says so; it does not fail. Within a run, keep pytest -n 2 at most.
Where the docs go: the commands go into CONTRIBUTING.md or the hub's dev docs, with no machine names, user names or paths outside the repo. A context/process.md entry gives the why (SQLite hides PG errors; parallel runs caused reaps).
Not: a change to CI.