Add bucketing/culling options to `sonalyze top`

[lars-t-hansen]

Depending on your point of view, adjacent records in the per-cpu load data stream can be merged or some records can simply be discarded: if we have three records, the middle can be dropped if we're interested in the average utilization between the first and the third, or alternatively, these records can all be used (somehow) to create an interpolated timeline with regular time steps that don't match the steps of the records themselves.

We might take inspiration from the bucketing and stepping options of the load and profile commands.