[jobsets] Fix bug in jobset atexit on local scheduler (#2312)

- Local scheduler was deleting succesfully completed jobsets. We avoid this now by ensuring that jobsets are killed or deleted when they are not running or waiting (in suspended state).

metaflow/plugins/kubernetes/kubernetes_jobsets.py

@@ -319,6 +319,8 @@ def _fetch_pod(self):
     def kill(self):
         plural = "jobsets"
         client = self._client.get()
+        if not (self.is_running or self.is_waiting):
+            return
         try:
             # Killing the control pod will trigger the jobset to mark everything as failed.
             # Since jobsets have a successPolicy set to `All` which ensures that everything has