Before the times would add up to a little over 58 seconds (assuming
the dtime steps keep up well enough); now it should add up to a
little over 60 seconds, getting us closer to an exactly 1 minute
sliding sample window.
Set the max bar width to 50% of the total accumulated time, so
there is always one longest bar unless 2 bars have 50% each, and
all non-longest bars are linearly comparable.
Increase accounting granularity so we have smoother expiration
of old entries.
