- lal-python --> python-lal
- lalinspiral-python --> python-lalinspiral

this fix removes the "wait time" where O(10^2s) of data was dumped until the whitener settled.

feature_extractor.py: remove unused imports, utils.py: switch feature dtype from double to single precision float

same as commit for segment file input, just with invert_output reversed in mksegmentsrcgate

currently making it the same procedure as when the buffer is non-gap but not enough samples (< autocorrelation_length), like itac

dump number of events in the coinc_events table.

feature_extraction: change block_stride for whitening, high pass from 1s to 0.25s, allow for a minimum downsampling rate to improve latency at expensive of computational cost

gstlal_feature_monitor: reduce number of missing channels messages, remove close connection line causing crashes upon shutdown

this enables to do multi-detector trigger generator & coincidence analysis
by inputs of channel-name, data, template, and PSDs from multiple IFOs.
In short, this is lalapps_StringSearch + ligolw_add + lalapps_burca of old pipeline.

like the old burca

__lt__() and friends

- use `Requires.private` in pkgconfig files to stop forcing downstream
  packages to link against everything upstream
- added `pkgconfig` as `BuildRequires` for RPMs
- bumped required version of pkg-config to 0.18.0

Makefile.gstlal_feature_extractor_offline: fix issue with generating segments from int8s conversion, add default safety info, clean up comments and unused variables

gstlal_feature_extractor_pipe: link feature combiner job to run after all feature extraction jobs are finished

gstlal_feature_extractor_pipe: fix some issues with changed imports, update params in offline makefile

binjfind needs them to identify sim_burst <-> sngl_burst coincs

Makefile.gstlal_feature_extractor_online: update kafka server params, ability to run off of /dev/shm easily

feature extraction: change trigger_time -> time, remove nan rows, specify true spans of datasets to comply with gwtrigfind

Set_caps seems to be executed *after* start, so in the previous commit audio_info will not be updated before the safety check and stumbles there. After this fix things seem to be going, and a sane value for chisq is obtained.