When we remove recipient nodes in the shard split fixture we might have an error in `replSetReconfig`. We can retry the reconfig as we use `with_naive_retry` but we do not fetch an updated config from the primary because replSetGetConfig is in a separate retry loop. This can lead a situation we can't escape from (if the config we try to apply is state).

Instead, in case of issue, we need to fetch the new primary, fetch the most recent config and reconfig using it.