Pymc4 upgrade, io functionality, greiner-quinn

[karink520]

This branch has several larger changes on it, which are:

• updates pyei to run on pymc v4 instead of pymc3. The sampling now runs on pymc.sampling_jax.numpyro_nuts, which results in a big speedup. Note that
• includes an implementation of a Gibbs sampler for the Greiner-Quinn model from
  James Greiner, D., and Kevin M. Quinn. "R× C ecological inference: bounds, correlations, flexibility and transparency of assumptions." Journal of the Royal Statistical Society: Series A (Statistics in Society) 172.1 (2009): 67-81. Although see issue Add tuning of width of proposal distributions in MH step within Greiner-Quinn method #98 for additional tuning that would be good to include. The two Wakefield model variants have been removed for the time being, because the Greiner Quinn method is a generalization, and because there seems to be a bug when the Wakefield models are sampled with pymc v4
• new io functionality for saving and loading fitted pyei models to disk.
   

[chanelrichardson]

Lots of changes and looks great! We've used these updates in Texas, and worked pretty closely with the new io_utils helpers, and have looked at the new version of diagnostics, and of course used the output summaries and vizzes, so I think that covers a lot of "testing" that could be done for this PR.