fix(task): prevent duplicate Slurm job submission on backend restart

RecoverStuckTasks now skips tasks that already have a slurm_job_id,
and ProcessTask adds a guard before the submitting step to prevent
re-submission even if a task is incorrectly re-enqueued.

Also deprecates POST /api/v1/jobs/submit endpoint (replaced by POST /tasks)
and comments out related handlers and tests.

internal/service/task_service.go

@@ -263,7 +263,15 @@ func (s *TaskService) ProcessTask(ctx context.Context, taskID int64) error {
 		}
 	}
 
-	// 13-14. Set ready + submitting
+	// 13-14. Set ready + submitting (guard: skip if already submitted to Slurm)
+	if task.SlurmJobID != nil {
+		s.logger.Info("task already has slurm job, skipping submission",
+			zap.Int64("task_id", taskID),
+			zap.Int32("slurm_job_id", *task.SlurmJobID),
+		)
+		return nil
+	}
+
 	if err := s.taskStore.UpdateRetryState(ctx, taskID, model.TaskStatusReady, model.TaskStepSubmitting, 0); err != nil {
 		return fail(model.TaskStepSubmitting, fmt.Sprintf("update status to ready: %v", err))
 	}
@@ -694,6 +702,13 @@ func (s *TaskService) RecoverStuckTasks(ctx context.Context) {
 		return
 	}
 	for i := range tasks {
+		if tasks[i].SlurmJobID != nil {
+			s.logger.Info("skipping stuck task recovery, already in slurm",
+				zap.Int64("taskID", tasks[i].ID),
+				zap.Int32("slurm_job_id", *tasks[i].SlurmJobID),
+			)
+			continue
+		}
 		_ = s.taskStore.UpdateStatus(ctx, tasks[i].ID, model.TaskStatusSubmitted, "")
 		s.mu.Lock()
 		if !s.stopped {