Choose correct `wfg_domain` during importance sampling

[stephengreen]

When we update data conditioning settings during importance sampling, we need to carefully update the domain objects.

There are two relevant domain objects in Dingo:

• The waveform generator domain is used first, when generating waveforms.
• The data domain is used when evaluating the likelihood. Waveforms generated in the wfg_domain must be suitably truncated, e.g., if f_max is reduced. For most updates to data conditioning, we update the data domain. However, to maintain compatibility between the domains, we sometimes need to update the wfg_domain. This PR sets wfg_domain.delta_f to match that of the data domain.

Previously, we were simply using the data domain directly whenever making an update to data conditioning during importance sampling. This is clearly wrong, so it has been corrected.

Closes #222

@hectorestelles Please test and confirm that the problem in #222 is resolved.

[hectorestelles]

This PR seems to solve the issue found in #222 , I could successfully run SEOBNRv5PHM with f_max=972.0

[stephengreen]

To clarify the behavior now:

• If duration is specified in the INI file, we update delta_f = 1 / duration for the data domain and the waveform generator.
• If minimum-frequency or maximum-frequency are specified, we update corresponding quantities in the data domain.
• If tukey-roll-off, sampling-frequency, or duration are specified, we use these in conditioning the data and update the window factor.
• If approximant is specified we raise an error.

Future to-do:

• Add functionality to update other waveform settings, i.e., approximant, generation minimum and maximum frequencies, reference frequency, and starting frequency. See #226.

@hectorestelles Once you're happy with this, please approve the PR.