Cumulative statistics for success ratio with higher weight given to recent data but without storing any history.
Clients call success() or failure() when an event occurs, and the ratio of success to total events
is accumulated. Older values are given exponentially smaller weight, with a decay factor determined by a duration
chosen by the client. The rate measurement weights decay in two dimensions:
in time according to the lapse period supplied: weight = exp((t0-t)/T) where t0 is the
last measurement time, t is the current time and T is the lapse period)
per measurement according to the lapse window supplied: weight = exp(-i/L) where L is
the lapse window and i is the sequence number of the measurement.