update to cleaner code with `arviz.extract`

[drbenvincent]

Lines like this represent a common pattern across many notebooks in the repo:

unpooled_post = unpooled_trace.posterior.stack(chain_draw=("chain", "draw"))

This is basically some xarray stuff to get a 'flat' vector of MCMC samples by collapsing over chain and draws. But it is arguably unclear, opaque, and potentially frustrating to many users.

I propose we use the new(ish) arviz.extract function instead. So we would end up with something like this, which is much much clearer to a wide range of pymc users.

unpooled_post = az.extract(unpooled_trace, group='posterior')

Tagging @twiecki and @OriolAbril for any thoughts

[twiecki]

Yes, way way better.

[OriolAbril]

I would only use group argument when we want to get variables that are not in the posterior group. We set this default because getting variables from the posterior is the most common operation and it is convenient to skip it