Optimize process of adding etcd members to the cluster

[invidian]

Following etcd documentation, if new member is being added to the cluster, we first do API call to etcd cluster to inform it about the new member, then we create new containers.

However, the time slot between those 2 operations will cause an outage, if we add more members than quorum currently consist of (e.g. if you run 1 member and you add 1 member of if you run 3 members and you add 5 members). As creating containers may take a bit of time, depending on node connection speed and disk speed, it would be better to configure and create member containers first, then add new members, then just start the containers.

This could be added as a PreStart hook for container creation or we could globally only get all containers to created state. This way the impact will be greatly limited.

[invidian]

Also adding a member as a learner and then promoting it in PostStart hook once it catched up would be good.