Metric Type for high variance gauge

The current implementation of the collector does a periodic snapshot every second of the metric store to the metric pipeline. This sampling work well with counter and gauge values since it give a high level view of the system.

But this feature doesn't work if we want to keep track of response time to a remote service like HTTP calls to elasticsearch, in this case we need a structure that keep a moving windows of the values, this could also be used for gauges to have more granularity.

Start by tracing the collector's periodic snapshot from the metric store into the metric pipeline. Define what moving-window behavior is needed for remote-service response times and higher-granularity gauges, then confirm the design supports both use cases and preserves the existing counter and gauge view.

Written by the indexing model from the issue text.