Refactor compute node stats as a resource tracker extension

Compute nodes record a number of statistics about instances they host, such as number of instances and number of each os type, and pass this information back to the scheduler. This information is stored in a table called compute_node_stats as attribute/value/compute-node triplets. The intention was that the compute_node_stats table would be extensible as any attribute/value pairs could be stored, but it introduced a join between compute_nodes and compute_node_stats that is expensive to process.

BP extensible-resource-tracking introduces a new text field to store json encoded information maintained by the resource tracker. There is an opportunity to refactor the compute node stats code to take advantage of extensible resource tracking. This would eliminate the compute_node_stats table and "normalise" how information is tracked at the compute node.