improvement(cluster): setup nodes in parallel

Before this change we were running 'node_setup' methods serially.
In majority of cases the time gap is not critical.
But there are cases such as setting up hybrid disks on GCE
where setup of disks on a single DB node takes more than 20 minutes.
And we have setups with 6 nodes.

So, mitigate this waste of time and money by parallelizing the
setup of DB nodes and do serially only startup where we have
limitation for bootstrap stage.