feat(sqlite): refactor orders schema + dashboard period filter

Replace import_orders (insert-per-run) with orders table (one row per
order, upsert on conflict). Eliminates dedup CTE on every dashboard
query and prevents unbounded row growth at 4-500 orders/sync.

Key changes:
- orders table: PK order_number, upsert via ON CONFLICT DO UPDATE;
  COALESCE preserves id_comanda once set; times_skipped auto-increments
- sync_run_orders: lightweight junction (sync_run_id, order_number)
  replaces sync_run_id column on orders
- order_items: PK changed to (order_number, sku), INSERT OR IGNORE
- Auto-migration in init_sqlite(): import_orders → orders on first boot,
  old table renamed to import_orders_bak
- /api/dashboard/orders: period_days param (3/7/30/0=all, default 7)
- Dashboard: period selector buttons in orders card header
- start.sh: stop existing process on port 5003 before restart;
  remove --reload (broken on WSL2 /mnt/e/)
- Add invoice_service, E2E Playwright tests, Oracle package updates

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

api/requirements.txt

@@ -8,3 +8,5 @@ apscheduler==3.10.4
 python-dotenv==1.0.1
 pydantic-settings==2.7.1
 httpx==0.28.1
+pytest>=8.0.0
+pytest-asyncio>=0.23.0