Feature: Window based aggregations

[Max Kühn]

Since sensor readings are often not smooth, it would be helpful to offer aggregations per time window:

# use the maximum pcpu of the last two intervals
--pcu=window(interval*2, pcpu, max)

A window should offer setting the window size, metric expression and aggregation type.

Open questions:

• What argument order to use? It seems natural to have time, metric[, type], especially if metrics can be long; but if there is a natural time window then metric[, time][, type] may be simpler.
• What aggregations to offer? At the top of my head, I can think of max, min, avg and perhaps a weighted/exponential average.