input source raw lookback cache for improved new policy responsiveness

Currently when a new policy is created, it takes up to 1 minute to collect metrics before being visible in a central system like Prometheus. We could improve this responsiveness (at the expensive of memory) by holding a small buffer of raw input events - for example, 2-5 minutes. Then when a new policy is created, in addition to collecting live metrics, it can can concurrently run against the look back data and generate immediate metrics for collection and dashboarding.