CI: pytest jobs fail on a fresh resolve because SQLAlchemy 2.1 resolves postgresql:// to psycopg 3 #38

Symptom

Both pytest (app) and pytest (migrations) fail at alembic upgrade head on every fresh resolve since 2026-09-25:

File ".../sqlalchemy/dialects/postgresql/psycopg.py", line 497, in import_dbapi
    import psycopg
ModuleNotFoundError: No module named 'psycopg'

First seen on #37 (run 36342478818). main last ran green on 2026-09-19, before the release below, and has not run since; the failure is independent of #37's change.

Cause

• uv.lock is not committed, so CI resolves fresh. SQLAlchemy 2.1.1 was uploaded 2026-09-25 and pyproject.toml declares sqlalchemy[asyncio] with no upper bound, so CI now gets 2.1.
• SQLAlchemy 2.1 changed the default PostgreSQL driver: postgresql:// now means psycopg (3), no longer psycopg2 (migration notes: "Default PostgreSQL driver changed to psycopg (psycopg 3)").
• app/settings.py:60 builds the sync DSN for Alembic with .set(drivername="postgresql"), so migrations ask for the default driver, and the project depends on psycopg2, not psycopg.

Options

1. Name the driver: drivername="postgresql+psycopg2". Smallest change, pins the meaning of the URL regardless of SQLAlchemy's default.
2. Move to psycopg 3: replace psycopg2 with psycopg[binary] and keep postgresql://. Aligns with SQLAlchemy's direction; larger change.
3. Cap sqlalchemy<2.1 until either of the above. Buys time only.

Either way the bare sqlalchemy[asyncio] dependency is what let a release change behaviour under CI; a lower bound and, for an application, a committed lock would have made this a deliberate upgrade.