feat(notifications): persistent notification center

- notifications table (goose migration) + sqlc queries; list with unread
  count, mark-one, mark-all endpoints under /notifications.
- Deployment success/failure now writes a notification row for the owner
  (insertUserNotification — best-effort, never breaks the deploy flow).
- Shell bell reads the real persisted feed: unread badge, click-to-read,
  mark-all, 30s polling. Replaces the transient builds+audit derivation.

Verified e2e: migration applies clean, list/read/read-all round-trips.

Authored By: TDvorak <[email protected]>

app/backend/internal/api/deployments.go

@@ -4,6 +4,7 @@ import (
 	"containr/internal/database"
 	"containr/internal/deployment"
 	"context"
+	"fmt"
 	"net/http"
 	"strings"
 	"time"
@@ -368,12 +369,19 @@ func runDeploymentAndSync(
 					`UPDATE services SET status = 'running', updated_at = $1 WHERE id = $2`,
 					time.Now(), service.ID,
 				)
+				insertUserNotification(db, userID, "deployment", "Deployment succeeded",
+					fmt.Sprintf("Service %s is now running.", service.Name), "service", service.ID.String())
 				return
 			case "failed":
 				_, _ = db.Exec(
 					`UPDATE services SET status = 'failed', updated_at = $1 WHERE id = $2`,
 					time.Now(), service.ID,
 				)
+				body := fmt.Sprintf("Service %s failed to deploy.", service.Name)
+				if current.Error != "" {
+					body = fmt.Sprintf("Service %s failed to deploy: %s", service.Name, current.Error)
+				}
+				insertUserNotification(db, userID, "deployment", "Deployment failed", body, "service", service.ID.String())
 				return
 			}
 		}