multiplying calibration factor in ROQ likelihood

- fix minor bug in multiplying calibration factor in roq.py
- add unit test of likelihood accuracy with non-zero calibration uncertainties
- The ROQ bases used in unit test are updated to ones trained against non-zero
  calibration uncertainties.
- The number of cases tested in TestROQLikelihoodHDF5 is reduced to reduce the
  run time of unit test.
- The error torelance of the unit test is increased as the error torelance of the
  new basis is larger than the previous one.

The `generate_component_masses` function generates component masses
from any pair of masses with sufficient information to do so,
`generate_mass_parameters` now takes any pair of mass parameters with
sufficient information and adds all of the mass parameters.

bilby/gw/likelihood/multiband.py and test/gw/likelihood_test.py: fix calibration errors in multiband likelihood and add tests of multiband likelihood for non-zero calibration errors

test/gw/source_test.py: add test for binary_black_hole_frequency_sequence and binary_neutron_star_frequency_sequence

Ethan Payne authored 4 years ago and Moritz Huebner committed 4 years ago

removed the likelihood matrix calculation within the likelihood function, and instead use a loop in the post-processing script

There seems to be an issue with colorbars in matplotlib 3.3.0.
As this code is not reguarly used, I'm removing it from testing to avoid
breaking scheduled tests.