Fix DBAPIError: use DateTime(timezone=True) and timezone-aware defaults in database models

Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Agent-Logs-Url: https://github.com/christianlouis/pop_puller_to_gmail/sessions/608dd6ae-4c0c-4076-a2e3-dec62c9923f3

CHANGELOG.md

@@ -8,6 +8,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Fixed
+- Fixed `sqlalchemy.exc.DBAPIError` raised by asyncpg when inserting timezone-aware `datetime.now(timezone.utc)` values into timezone-naive `DateTime` (TIMESTAMP WITHOUT TIME ZONE) columns: changed all `DateTime` column definitions in `database_models.py` to `DateTime(timezone=True)` (TIMESTAMP WITH TIME ZONE) and replaced all `default=datetime.utcnow` callable references with `default=lambda: datetime.now(timezone.utc)` for consistent, timezone-aware timestamps throughout the ORM.
 - Fixed `ProgrammingError` (`cached statement plan is invalid`) raised by the asyncpg dialect during startup: SQLAlchemy's asyncpg wrapper maintains an LRU prepared-statement cache per connection (default size 100). When `Base.metadata.create_all()` executes `CREATE TYPE … AS ENUM` DDL inside a transaction, PostgreSQL invalidates the cached plans for that connection. The next enum-type existence check then fails because the dialect tries to reuse the now-stale prepared statement. Fix: set `prepared_statement_cache_size=0` in `connect_args` on `create_async_engine` to disable the cache entirely, which is the documented SQLAlchemy recommendation for DDL-at-startup scenarios.
 - Fixed `UndefinedTableError` on first boot: the lifespan startup event now calls `Base.metadata.create_all()` via the async engine before attempting to seed default settings, so all tables are created automatically when the database is empty (e.g., fresh PostgreSQL container with no Alembic migrations run yet).
 - Fixed frontend API calls being hardcoded to `http://localhost:8000` in production: `NEXT_PUBLIC_API_URL` is baked into the JavaScript bundle at Next.js build time, so it can never be overridden at container runtime. Replaced the `NEXT_PUBLIC_API_URL` mechanism with a Next.js Route Handler proxy at `/api/v1/[...path]` that reads `process.env.BACKEND_URL` at server startup and proxies all `/api/v1/*` requests to the real backend. The frontend Axios client now uses a relative base URL (`/api/v1`), which also eliminates the CORS issue since the browser only ever talks to the same-origin Next.js server. Update `BACKEND_URL=http://backend:8000` in `docker-compose.new.yml` (or your deployment env) to point the proxy at your backend.