docs: add a tutorial / how-to-guide for using pyro with sbi

[janfb]

We now can use sbi estimators as pyro models (#1491 ) and it would be nice to have a short guide how to do this.

This could be integrated into the tutorial on mnle or trial-based inference (https://sbi.readthedocs.io/en/latest/tutorials/12_iid_data_and_permutation_invariant_embeddings.html)

or a FAQ-like how-to-guide in the form of a code snippet.

[sethaxen]

Is there a place where it would make sense to also document use of a learned posterior as a prior in a Pyro model?

The killer app would be to demo use of a learned likelihood in a nontrivial hierarchical model, which sounds like it might be worth its own tutorial.

[sethaxen]

As we discussed, a necessary but not sufficient check that things worked well when (specifically, that the learned likelihood is sufficiently accurate in the region of parameter space where Pyro's posterior concentrates) using a learned likelihood in Pyro would be to check that the joint posterior/posterior predictive draws from the Pyro model (generated using Predictive), are consistent with the draws obtained by running the actual simulator conditioned on Pyro's posterior draws.

[janfb]

Is there a place where it would make sense to also document use of a learned posterior as a prior in a Pyro model?

The killer app would be to demo use of a learned likelihood in a nontrivial hierarchical model, which sounds like it might be worth its own tutorial.

In the long run we can add a longer tutorial in Examples showing the general full workflow using a non-trivial hierarchical model, e.g., the "full DDM", where we would cover all those cases, including the predictive checks etc.