aggregator execs + ll_inspiral_pipe: allow more control with number of routes processed per job, increase number of aggregator jobs in online dag for scalability

gstlal_ll_inspiral_pipe: rename output dir for gstlal_ll_dq for consistency, create multiple gstlal_ll_inspiral_state jobs to handle aggregation better

- 14cf8072 introduced a ln L offset for singles but put it into the numerator.  that is not allowed:  the numerator (and the denominator) must be the *real* distribution density for signal (and noise) candidates, we rely on that being true for astrophysical rate estimation (and false-alarm probability measurement).  ranking statistic hacks go into the ranking statistic, which we define based on these two PDFs in far.py

- make the exclusion of sub-threshold triggers from the kwargs_from_triggers() method permanent, and put an assert into .__call__() to make sure no sub-threshold SNRs leak in.

--event-chisq to plot an snr and chisq on the snr chisq plots

- allow the threshold to be adjusted manually (default is the current
  behaviour, which is to set the threshold from the lower bound of the
  extinction model's domain of validity)

- weren't used for anything, adds baggage to the code.  in the future, it
  might be worth adding an option to select which offset vector to process,
  but none of what this patch removes would be used for that

- a bunch of code was added related to the alphabet soup approach, and then
  commented out.  this patch removes it.  it can be reverted later if
  somebody wants to resurrect it

- define a stand-alone method to implement the ranking statistic's
  fast-path cut, and use that method for both the cut itself *and* the
  coinc sieve passed to thinca to ensure the two are always identical.
- this is the first step in making this cut much *MUCH* more agressive.

chad.hanna authored 6 years ago and Kipp Cannon committed 6 years ago

- add option to cluster after coincidence. Note the biggest SNR and biggest LR (if available) trigger is kept
- fix bug in max over lr and snr
- optionally limit the far of single candidates by live time for online analysis
- make online 'clustering' before LR computation
- fix clustering

Patrick Godwin authored 6 years ago and Kipp Cannon committed 6 years ago

consolidate DAG classes from inspiral_pipe.py + dagparts.py (e.g. generic_node) into dagparts.py, move non-specific inspiral functions from inspiral_pipe.py to dagparts.py, changes reflected in several dag generation execs