Host discovery should be faster

Since some deployments will want to run host discovery periodically, we should make sure that the process is as lightweight as possible.

The "discover_hosts_in_cells_interval" config option was added in Ocata which is disabled by default but allows smaller deployments to run a periodic task from the nova-scheduler process which discovers new hosts and maps them to a cell. This is a way to automate the "nova-manage cell_v2 discover_hosts" command which needs to be run after new compute nodes are added to the deployment.

In the initial form in Ocata, the periodic is not as efficient as it could be. It queries all cells and then for each non-cell0 cell, it queries all compute_nodes records and for each compute node record, it checks for a related host_mappings record. If a host mapping is not found, one is created, else it just continues.

The point of this blueprint is to make this operation more efficient by flagging when a compute_nodes record is mapped to a host in the nova_api database, and then we can filter those compute nodes out of the initial query per-cell when running discover_hosts.