Slice (and thus round) duration should be based on wall time

[alistairking]

Currently the slice timer is fixed to the ideal duration, but it would be good to dynamically adjust these times to allow the prober to dump timeseries stats as close to the end of the actual interval as possible.

For example, currently if the prober starts running at 10:05am, it will collect data until 10:15, and then dump that as the 10:00 -> 10:10 interval, effectively meaning that data will be 5 min late. If we could dynamically adjust the slice timer so that each round finished within a few seconds of the timeseries interval, we would have better (lower) latencies, and better alignment amongst probers.

[alistairking]

I'm no longer convinced that this is a problem. We should add some metric that tracks the "realtime delay" of the prober. I.e. the delta between the wall time and the interval time when metrics are dumped.