Add a new stats collector component that gather summary of node and pods usage from each node. This information can then be used by components like scheduler and autoscaler.

First version will expose node usage summary for cpu and memory. To start with, summary includes stats for average, maximum, and 90 percentile usage in last minute/hour/day. The stats are precomputed in cAdvisor and routed to the collector through kubelet.

@davidopp @bgrant0607