Clarify durations and speeds in flow slices

Currently the relations between durations and speeds in a flow slice and statistic is a bit fuzzy and messy concept. Likely the flows can even overflow into a neighboring time slice.

We probably want this:

• Use in-aggregator times instead of guts times for all the intervals, times, etc, so we can get them at any time.
• Start and end of the flow slice. These would be either the start and end of the flow itself (if it starts and ends in the current slice) or the ends of the slice.
• Duration used for average speed computation (while the start and end may be used for computation of real duration, we want to base max speed computation on the whole length of the slice and we don't want the average to be higher).

The start and end are min/max over the combined statistics. The duration is summed up together like now.

Also, the flows and times should be independent in each direction.