Caching host aggregate metadata for scheduler

Many scheduler filters replying on host aggregate metadata for placement decisions. However, it is a known issue that DB query in filter has a performance. Because for each scheduling request, there gonna be roughly n(hosts) x n(filters has aggregate meta query) DB calls generated, which is very unscalable for any large deployment.

This blueprint is proposing a caching layer for aggregate metadata queries in scheduler. It will be based on underlying aggregate_host_get_by_metadata_key DB api to retrieve all host metadata for a specific key, if accessed, in one batch. If not metadata key not specified and only query by host, the actual cache fetching will be delayed to the key accessing time.

This blueprint will also add additional config options to control whether to enable this feature (default off), and the timeout value for the cache.