Bayesline

Total rewrite of BayesLine initialization/burn in phase. Will make BW runs reproducible (the BL burn in and prior-setting steps are now deterministic instead of stochastic), and will improve runtime, dramatically-so for large TFVs.

PSD setup for GW170817-like L1 data (with the monster glitch) will still require human intervention.

This is the extent of the planned changes for using BW for spectral estimation as input into CBC PE. Testing/benchmarking/review prep. for that application can begin now.