Improvements Weather Example

Followup of https://github.com/pymc-devs/pymc-extras/pull/702

From Ricardo:
> I'm curious (not asking to incorporate in thr example) what would happen with a fixed probability model (would increase posterior of sigma?). And whether we could infer the discrete change point day by marginalizing it as well. That may fail/ not be yet supported

I wanna explore this once the PR is merged.