Change Broker Allocation Algorithm

[kthare10]

https://fabric-testbed.atlassian.net/browse/FIP-714
Change Broker Allocation Algorithm to spread resources across available hosts more evenly as opposed to first fit.

[kthare10]

OOM issue has been resolved by reserving 24GB of RAM for the host and allocating rest to CF for allocation: https://fabric-testbed.atlassian.net/jira/servicedesk/projects/FIP/queues/custom/44/FIP-1405

Following change is requested for Allocation to avoid users from exhausting all shared NICs on a specific host.

Maybe we should have something like: Choose the first host in the list where the VM fits and the host has at least X% of the cores/ram/disk/shared_nics available, unless all the hosts have less than X% of the cores/ram/disk/shared_nics available. If all the hosts have less than X% of the cores/ram/disk/shared_nics available, then choose a random host that has enough resources.