- increase to 32 seconds for consistency with the corresponding default
  value in gstlal_inspiral's command line options

- don't rely on singles being flushed to disk to do this for us, takes too much memory to do so

- don't recompute coincidence_back_off in run_coincidence(), only compute when XML document changes (performance improvement)
- expose discard boundary calculation for use by calling code

- remove hard-coded 1 s end time offset now that this has been fixed in itac
- see 909f633e and related patches
- add assert to check that no trigger's end time lies outside of the buffer in which is is found

PyList_Append increments the item's reference count.

gstlal_compute_strain:  Added latency to firbank (only applies when using --factors-from-filters-file) in kappa demodulation routine so that pipeline doesn't hang

- a variety of import related adjustments and fixes

The old approach of storing a pointer to the COMPLEX8TimeSeries in
the GSTLALSnglInspiral struct did not work because we would often
add more than gst_buffer_get_max_memory() memory objects to a buffer.
Consequently, GStreamer would merge the memory objects together,
deallocating the original objects and causing the COMPLEX8TimeSeries
to be freed too early. There is no safe way to use a GstBuffer to
store a pointer to out-of-band auxiliary data.

The new approach is to store the SNR time series samples inline in
the buffer.

lal_smoothkappas: Now, the order of elements in the average array is insensitive to the start time and the array size. This should resolve the remaining discrepancy in h(t) computed with different start times.

Changed tagname to 'pe'

gstlal_compute_strain:  Removed audiocheblimit element from pipeline and used lal_firbank in its place (audiocheblimit appears to have a bug)

changed 'max likelihood' to 'mode' for plot

added assertion for when database files not in directory

compute log posterior with rate_estimation.LogPosterior

- fix install paths for python modules in debian packaging scripts

- from glue.iterutils.Highest (getting deleted)

gstlal_compute_strain:  Added DARM line coherence gating to kappa_tst calculation.

Adding Atlas linking flags for FFTW; wasn't there before, but I haven't seen a performance difference with/without it.