Wait for replication in shardingtest.restartAllConfigServers and shardingtest.restartAllShards to ensure that a node won't enter rollback after. This can cause problems if the action a test takes after the restart caches connections, because they'd be closed, e.g. if we call stopSet immediately after.