Re-read a delivery before failing it for a missing target

failMissingTarget failed the delivery as the batch had read it, so a
delivery a worker sent and let go between the batch read and the
ownership check could end failed after a successful attempt. It now
reads the row once it owns the delivery and fails it only if the
status is still the one the batch read, as processNewTask does. A row
that cannot be read is left alone.

TestSweepPending_TargetDeleted now checks that the first sweep already
queues the healthy delivery and the second does not queue it again.

Model: opus-5-5

internal/delivery/engine.go

@@ -1270,6 +1270,30 @@ func (e *Engine) failMissingTarget(
 
 	defer e.inflight.release(d.ID)
 
+	// The batch was read before ownership was taken, and a worker may
+	// have settled the delivery and let it go in between. Only a row
+	// still in the status the batch read is failed.
+	row, err := e.loadDelivery(webhookDB, d.ID)
+	if err != nil {
+		e.log.Error(
+			"failed to load delivery",
+			"delivery_id", d.ID,
+			"error", err,
+		)
+
+		return
+	}
+
+	if row.Status != d.Status {
+		e.log.Debug(
+			"delivery already handled, not failed",
+			"delivery_id", d.ID,
+			"status", row.Status,
+		)
+
+		return
+	}
+
 	targetType, reason := e.missingTargetReason(d.TargetID)
 
 	e.log.Warn(