Provide helper utilities for blinding by region

It would be nice to provide a way to ask or generate the data of a particular model with some regions blinded. Blinding would mean that, instead of using the associated observations, one uses the total yields of MC in that region (pre-fit).

I could see this, maybe as a pyhf.contrib or pyhf.utils API, but unsure of which. It would be something that consumes a pyhf.pdf.Model and observations, as well as a list of regions to blind, and returns the corresponding yields.

Some other ideas:

• can help generate asimov data
• blinding should possible round the data to integers (to represent the fact that the observations are always integers – morally?)
• can include an option to ignore the signal sample(s) when generating the blinded yields for a region

[alexander-held]

related issues: https://github.com/scikit-hep/pyhf/issues/1106, https://github.com/scikit-hep/pyhf/issues/1029

[kratsg]

To control the unconstrained NPs doesn’t give you region-by-region control though?