Tweaks to make to the Geweke test

[njtierney]

We run the Geweke test to check that greta recovers the correct distribution. We have run in to some issues with this, potentially related to TF1 - @goldingn notes that this might be some of the issue:

to make the Geweke test run fast enough, the code I wrote for TF1 does some open-heart surgery on greta internals:
https://github.com/greta-dev/greta/blob/master/tests/testthat/helpers.R#L686
which likely does not work any more with the new internals. I don't think the data list object is being reused each time now, so x isn't being updated. That would cause the Geweke test's Markov chain to wander off, and produce a plot just like this

[njtierney]

This would resolve #606

[njtierney]

Fixed in #734