Adding tidal waveforms

[ThibeauWouters]

This pull request tries to merge the new ripple waveforms TaylorF2 and IMRPhenomD_NRTidalv2 waveforms into the jim main branch. I'd like to give a few more comments about this:

For source code: besides adding a way to interface with ripple to get the waveforms, I have made small changes to likelihood.py to enable users to use a different waveform as a reference waveform in the heterodyned likelihood. This is necessary for the IMRPhenomD_NRTidalv2 waveform, where we want to remove the Planck taper from the waveform as explained by our paper.

For the examples: these are not finalized yet, for the following reasons:

• I am using my custom plotting script to eyeball the final results of the runs and make diagnosis plots. If this might be useful for other jim examples as well, I can clean it and put it in the example directory.
• For the GW170817 runs, I am using the dataset that I used for the BNS paper to cross-check the results and to avoid having to deal with the glitch in the data when fetching the data. Therefore, the script will throw errors unless users run it on CIT.
• I have trouble reproducing the results from the BNS paper, possibly because of the version changes in flowMC and jim. I am wondering if it might be that one has to re-tune the hyperparameters because of the version change?

[kazewong]

Is there a reason why you would use another waveform model as the reference waveform?

[ThibeauWouters]

The NRTidalv2 waveforms have a tapering window for the amplitude such that the amplitude reaches zero at around 1.2 times the merger frequency, which depends on the parameters of the waveform. We observed that this breaks the relative binning in case the merger frequency of the reference waveform is lower than of the waveform of sampled parameters, since the reference waveform enters the denominator of the relative binning approximation and will cause it to blow up. We fixed this and improved the stability by removing this taper from the reference waveform while still using the taper for sampled waveforms. Our pp plot showed we still get decent and consistent results, but this therefore requires a bit more flexibility in the likelihood file.