Update to GWPopulation breaks search_keys

This line in data_analysis.py infers the parameters that each model in the hyper model expects. However, this commit to gwpopulation removed these parameters explicitly from the __call__ for mass and redshift classes and assumes they are kwargs. This means that getfullargspec (called by infer_args_from_function_except_n_args doesn't actually see the parameters that the model needs, as they are no longer explicitly listed in the inputs. The priors for those parameters that are missed from search_keys then get deleted in this loop.

Since the mass and redshift models now have the variable_names property, we could just replace the first line linked on this issue with a modified version of this function in bilby hyper model that just returns param_keys (note I think we will need to keep the KNOWN_ARGUMENTS part of the loop because of the parameters of the spin functions not being exactly what is sampled).