Celery worker entrypoint doesn't load .env; connects to SQLite with module schemas #94

Summary

The documented worker command — uv run celery -A scripts.run_worker:celery worker -l info — doesn't load host/.env, so Settings() falls back to the SQLite default and the model __table_args__ keep their per-module schema qualifiers (datasets.datasets_dataset etc.). When a task tries to execute against a real Postgres database (because the broker is reachable but the worker's settings disagree with the web process), every query fails with either:

• sqlite3.OperationalError: unknown function: now() (when the worker is actually using SQLite), or
• psycopg2.errors.UndefinedTable: relation "datasets.datasets_dataset" does not exist (when the worker is wired to Postgres but the model still has the schema-qualified table name from _default_provider).

Repro

1. Set SM_DATABASE_URL=postgresql+asyncpg://... in host/.env (which uvicorn picks up automatically through pydantic-settings via cwd).
2. Start the worker the way host/Makefile's `worker` target does: `uv run celery -A scripts.run_worker:celery worker -l info`.
3. Enqueue a real task (e.g. `lacowiki_datasets.convert_dataset`).
4. Worker fails to write `background_tasks_task_execution` rows because either the env var is missing or the schema provider doesn't match the dev migration.

The dev-only helper host/scripts/reset_admin_dev.py works around this with:

import simple_module_db.base as _smdb_base
_smdb_base._default_provider = lambda: _smdb_base.DatabaseProvider.SQLITE

…which is itself a sign the seam is too sharp for normal app code.

Expected

scripts/run_worker.py (or the build_celery factory it calls) should:

1. Load .env from the host directory before instantiating BackgroundTasksSettings/Settings.
2. Apply the same provider/schema convention the web process applies, so model __table_args__ line up with the actual database the worker reaches.

Workaround

Wrap the worker entrypoint:

# scripts/run_worker.py
from dotenv import load_dotenv
from pathlib import Path
load_dotenv(Path(__file__).resolve().parent.parent / ".env")

import simple_module_db.base as _smdb_base
_smdb_base._default_provider = lambda: _smdb_base.DatabaseProvider.SQLITE  # match dev migrations

from background_tasks.celery_app import build_celery
from background_tasks.settings import BackgroundTasksSettings
celery = build_celery(BackgroundTasksSettings())

Acceptance

• A fresh sm new --preset full host's worker command runs against the same database the web process uses, with no extra wiring in scripts/.