Incorrect uncertainties whe used with PyMC4 and PyMC5

[ivanistheone]

In case anyone is using this library with PyMC v4 or v5, note the need to "un-transform" the variables before computing the Hessian, otherwise the uncertainty of the estimates will be incorrect.

I read about this problem here: pymc-devs/pymc#5443 (comment)

Applying the fix to the quap function looks like this:
ivanistheone/pymc_Rethinking_3@0f498e7

Note this is not an issue in PyMC3 since the find_MAP and find_hessian are both base don un-transformed coordinates.

I'd be happy to submit a PR with this fix, though not sure how to make the code compatible with both PyMC3 and PyMC4+.