Update nested sampler integration

[johannesulf]

This PR improves the nested sampler integration. The API now allows more direct communication between the user and the sampler by passing keyword arguments directly to the sampler function calls. Prospector now looks at the signatures of function calls in the different sampler packages and dynamically assigns the appropriate keyword arguments passed by the user. At the same time, the following keyword arguments remain "global", i.e., prospector translates them into the corresponding keyword arguments for the respective sampler.

• nested_neff: The minimum effective sample size.
• nested_nlive: The number of live points.
• verbose: Whether to give progress updates.

I also ended up removing the keyword arguments nested_bound, nested_sample, nested_walks, and nested_dlogz. My reason was that it was somewhat confusing that they had slightly different names than in dynesty. Plus, they don't always have a corresponding parameter in the other samplers. The main practical advantage of the new implementation is that prospector doesn't needs to follow changing APIs in the samplers. For example, if a sampler implements a new keyword argument, users can directly utilize it without having to update prospector. Finally, whether parallelization with MPI works as expected for every sampler may still need to be checked.

[bd-j]

Thank you @johannesulf! I think the argument inspection will be very useful. I'll pull a version and make sure tests_sampler runs.

[bd-j]

So, I gave this a try with tests_samplers.py with minor changes, and it seems to run well. My only concern is this line gives a potentially huge number of not always helpful warnings, since in common practice the full kwargs dictionary often includes parameters related to other aspects of the code.

I understand passing non-relevant keywords is not great practice, but until it is improved in many disparate parts of the code I wonder if there could be something like an additional 'debug' or 'validate_sampling_keywords` variable that defaults to False to keep this line from running?

Happy to accept this PR and make such a change afterwards.

[johannesulf]

Yes, totally agreed. I tried out the code and also noticed the warnings. When coding this up, I worked under the assumption that only sampler keyword arguments would be passed here. Happy to have this line removed or adjusted according to your suggestion.