I don't see any actual failure in the logs of failed change_stream_shard_failover.js test except it spends too much time in some of the steps:

1. The `replSetStepDown: 300` command might be blocking the former primary re-election for too long. Perhaps try the timeout of 60 or 100
2. The `awaitNodesAgreeOnPrimary()` default timeout is 10 minutes. Perhaps wait for some smaller timeout. I see in the logs this step sometimes takes too long, maybe because of #1

I would also add better logging around those cases.