Refactor the logic around use of floating ips in node groups and clusters

Currently, there is a boolean in the configuration file "use_floating_ips" that means multiple things at the same time:

1) All instances must have a floating ip
2) Every node group therefore must supply a floating ip pool when using neutron, and in the nova case nova must be configured to auto-assign floating ips
3) Sahara must use the floating ip for management
4) These requirements are in force for every user, every node group, every cluster!

This logic should be broken apart. As we move toward a world where virtual and baremetal clusters (and even node groups) co-exist, we should have more flexibility in handling floating ips.

When using neutron, individual node gropus should be able to declare a floating ip pool or not. If use_floating_ips is set to True, then Sahara should use the floating ip for management of an instance *if it exists* but not require it.