Add in "level A" jobs

This changes the way multiple jobs are handled. Previously, they simply used the condor "Queue". Instead, this now creates a list of level_A jobs (I'm not 100% sure about using that term), each level A job has one data generation and at least one data analysis job which are then chained parent-child.

The result is that if one level A job fails, it only takes down that chain, not the whole DAG.

During the course of this change, I made several other fixes listed here:

1. bilby_pipe_generation now takes an argument --injection-file (previously it tried to infer it which was buggy and wrong)
2. generation and analysis now take an argument --idx which passes around which "job" they are in the level-A job list
3. create-output is made optional (running large number of jobs can cause race conditions as they all try to read the rcparams file)