Add AI-powered Bayesian inference (AI Priors/AI posterior)

[seanohagan]

Adds a Bayesian PPI variant AI-powered Bayesian inference via ppi_py/ai_priors.py. Exposes functions sample_ai_posterior and calibrate_dp_alpha_empirical_coverage_estimate. Currently the calibration function is based on a coarse, low-fidelity grid search as high-fidelity calibration can be computationally intensive for this method.

Something to note is that the implementation is general and the sample_ai_posterior requires users to specify a loss function which is Callable. For parallel computation on the CPU to work, the function loss must be pickleable (i.e. defined at the top-level of a module). To overcome this and allow for lambda functions/nested functions/etc may require an additional dependency like cloudpickle.

Adds tests via tests/test_ai_priors.py and an example notebook on the galaxies dataset via examples/galaxies_aip.ipynb.

Uses examples/aip_utils.py solely to add code to make a plot in the example notebook. Could be potentially refactored out, with the code moved to the notebook itself or the existing plotting function to be suitably modified.