Result saves DeltaFunction prior for runs with marginalization
When running jobs with time_marginalization and/or distance_marginalization on, when I access the prior saved int he result file it says it is a DeltaFunction at the reference value, as the likelihood sets it to be for distance here). I would expect the prior should be the user-input prior, as the reconstructed posterior reflects this prior. However, this is not the case for versions <= 1.1.5; in these earlier version, the result saves the input prior for marginalized parameters so some more recent change affects this behavior. This can be reproduced by running, for example, the marginalized_likelihood example with the current bilby release version and looking at the saved result.priors. I have not yet been able to track down where the issue is coming from, but it appears to be from something added in 1.2.0.