Support configurable metrics aggregations with percentile/histograms transformation

For aggregation in for. example statsd would be awesome to have in Vector transformation to do aggregations like min, max, stdev, mean, count, sum on a defined pipe of metrics but what is also needed is to have bucketing to get custom percentiles from a defined list to calculate like

percentiles = [50,70,90,95,97,98,99]

Also, histogram support with an example for ms buckets in duration calculation.

histogram = [1,10,100,1000,2000,3000,4000,5000,6000,7000,8000,9000,10000]

this will add a set of tags based on histogram per aggregated window of metrics. We may also have the option to choose if we like to add histogram and/or percentiles data as new fields or new tags list of buckets - Some systems better support fields query language based on some better query on tags at the end-user experience.

This is connected with https://github.com/timberio/vector/issues/676 to have defined aggregation windows per this transformation.

I think would be nice to move aggregations as most do from source more behind with more control from Vector. about which metrics we like to aggregate and which not.

For this example, we can have current statsd input with some improvement on supported types - https://github.com/timberio/vector/issues/2603 and move aggregation into transforming part to be more elastic. This may also apply for metrics from logs and more sources.

Start by reading the related aggregation-window issue 676 and the StatsD improvements issue 2603, then trace how the current StatsD input and transformation pipeline handle metrics. Done would require an agreed design and implementation for configurable aggregation windows, percentiles, histograms, and output as fields or tags.