feat(notifications): dedicate AI Failure Detection notification event (#1794)

Split Obico failure-detection dispatch out of the multiplexed
  on_printer_error event onto its own on_ai_failure_detection event so
  users can subscribe to AI alerts without also enabling HMS hardware-
  error pages, and so the discoverable label "AI Failure Detection" is
  what subscribes them rather than the unrelated "Printer Error" toggle.

  New column on notification_providers (default False, branched
  SQLite/Postgres migration), new notification_service.on_ai_failure_detection
  method, new ai_failure_detection template, obico_actions._notify swap.
  Frontend gets a summary badge, a toggle row with description, and ntfy
  priority surfacing. 14 new tests pin the routing + the regression guard
  ("Printer Error" alone must NOT receive AI notifications now). 11 locales
  covered.

  Existing providers keep working: HMS hardware errors continue to ride
  on_printer_error unchanged; users who want spaghetti alerts opt in via
  the new toggle.

backend/app/api/routes/notifications.py

@@ -47,6 +47,7 @@ def _provider_to_dict(provider: NotificationProvider) -> dict:
         # Printer status events
         "on_printer_offline": provider.on_printer_offline,
         "on_printer_error": provider.on_printer_error,
+        "on_ai_failure_detection": provider.on_ai_failure_detection,
         "on_filament_low": provider.on_filament_low,
         "on_maintenance_due": provider.on_maintenance_due,
         # AMS environmental alarms (regular AMS)
@@ -127,6 +128,7 @@ async def create_notification_provider(
         # Printer status events
         on_printer_offline=provider_data.on_printer_offline,
         on_printer_error=provider_data.on_printer_error,
+        on_ai_failure_detection=provider_data.on_ai_failure_detection,
         on_filament_low=provider_data.on_filament_low,
         on_maintenance_due=provider_data.on_maintenance_due,
         # AMS environmental alarms (regular AMS)