Change interpolation order for distance marginalisation
For loud events with distance marginalisation, the recovered distance and inclination posteriors are spiky, see attached image for an example.
I think we can fix this by just using cubic interpolation rather than linear interpolation used currently.
We can just pass it as a keyword argument at https://git.ligo.org/lscsoft/bilby/blob/master/bilby/gw/likelihood.py#L565.
The attached example is SNR ~ 100, we may want to see how far we can push this and whether we need to look at a different interpolation function.