- this patch replaces snglcoinc's coincidence engine with a native
  streaming engine.  the old implementation was written at a time when it
  was believed that time slides were the focus of coincidence analyses, and
  was optimized for performing very large numbers of time slides.  we now
  know time slides are not how good noise models are constructed and that
  it is better to optimize for high trigger rates so as to allow for the
  lowest possible thresholds.  that means optimizing not for high offset
  vector count but, instead, for ram use and lower comparison operation
  count, i.e., adopting a "streaming model", where events flow into the
  comparison engine and coincidences flow out, a relatively small number of
  time shift offset vectors are considered, and events are only held in
  memory until they are no longer needed to construct coincidences.