The SNR time series epoch has to take into account the original
template's time shift.

- the version inherited from dict returns a dict instead of an
  HorizonHistories instance, and makes references to the values in the
  original object so that modifications affect both the original and the
  copy.
- this override returns the correct type, with deepcopy()s of the values

- introduce .functional_integral() methods to both NearestLeafTree and
  HorizonHistories classes
- reimplement .weighted_mean() as a wrapper of .functional_integral().
  this results in a performance penalty since it must invoke
  .functional_integral() twice but avoids the code duplication that would
  result otherwise
- allow 0/0 in .weighted_mean().  if the numerator of the mean is
  identically 0 and the weighted denominator has also evaluated to 0 then
  return 0 instead of reporting a divide-by-0 error.  this is done
  presuming that this is caused by the weight function being defined in a
  way that causes it to go to 0 when the data being averaged goes to 0 and
  that if a proper limit of the ratio was evaluated it would be found to be
  0.

idq_aggregator.py: changed hdf saving scheme for datasets (saves multiple datasets per file keyed by channel and rate), hdf file saving for non-aggregates 2x faster, fixed bug with aggregation that caused missing triggers; gstlal_idq_trigger_gen: major refactoring in buf_handler and hdf saving functions

idq_aggregator.py: changed indexing scheme from MultiIndex to DatetimeIndex, multilevel aggregation fixed (except for possibly some edge cases)

- utils --> ligolw_utils

- introduce internal cacheentry class to hold entry information
- remove limit on cache size
- when initializing, move loop over instruments into .add_to_cache()
- add min_instruments parameter to the cache key
- in P_instruments_given_signal(), handle off instruments better
- record P(instruments | signal) in the SNRPDF cache
- enables generalization of ranking statistic away from fixed P(instruments | signal) marginalized over horizon distance history

- replace some generator expressions with lists that become generator
  expressions again in Python 3

- seems to be the source of the plague of lock-ups we've been suffering from

- *** REVERT AFTER O2 ***

tree.py: put prefactor to 1 and pad the equal mass line as well as remove mass scaling of deltas and increase metric tolerance