Numpyro Conversion Guide

## Tell us about it

Related to issue #1489.

Create a Numpyro conversion guide in the style of the [emcee guide](https://python.arviz.org/en/stable/getting_started/ConversionGuideEmcee.html).

This conversion guide would highlight a set of tasks of increasing difficulty for converting data from Numpyro to ArviZ, and include labelling event dimensions (see #2441).

This guide should also cover at least the following:

- Passing in a prior predictive
- passing coords
- inferring batch dims + event dims
- passing in posterior predictive
- passing in constant data
- (bonus) passing in prediction constant data and predicting an out of sample group

## Thoughts on implementation

The eight schools example [here](https://python.arviz.org/en/stable/getting_started/CreatingInferenceData.html#from-numpyro) can be extended to cover this.

For inferring event dims, the eight schools example could be reframed as `theta = mu + theta_offset` where `theta_offset ~ ZeroSumNormal(tau)` and theta is deterministic.

Start with the linked emcee conversion guide and the eight schools example in CreatingInferenceData, then review issue #2441 for event-dimension labelling. Extend the example with progressively harder Numpyro-to-ArviZ tasks covering prior and posterior predictive data, coords, batch and event dimensions, constant data, and optionally out-of-sample prediction. Done means the guide documents these conversions with runnable examples.

Written by the indexing model from the issue text.