add interesting validation case

[lukasheinrich]

Description

This PR does not modify any pyhf code

In the process of validating MBJ #217 we notices that roohf allows multiple systematics of different type to share the same nuis parameter (with I guess the implicit understanding that the structure of the nuis parameters must be compatible)

The primary example is a sample that has both

1. a HistoSys
2. a OverallSys

both are functions of a single nuisance parameter. with --qualify-names pyhf forces the names to be different even though they are the same in the XML, but roohf actually allows this.

This also adds a test script that runs a single point using roohf

The results between roohf and pyhf are the same if we use the config example_notshared.xml in roohf

i.e.

pyhf xml2json example/example.xml --qualify-names |pyhf cls
pyhf xml2json example/example_notshared.xml |pyhf cls

both produce the same result as

hist2workspace example/example_notshared.xml
python ../run_single.py example_combined_GaussExample_model.root

but is different from

hist2workspace example/example.xml
python ../run_single.py example_combined_GaussExample_model.root

Checklist Before Requesting Approver

• Tests are passing
• "WIP" removed from the title of the pull request

[coveralls]

Coverage decreased (-0.2%) to 97.345% when pulling 85a8ac5 on validation/shared_nuispars_across_types into 421ce3f on master.

[kratsg]

My only concern is the addition of binary files here.

[lukasheinrich]

we are already have some here

https://github.com/diana-hep/pyhf/tree/master/validation/xmlimport_input2/data

and they should all be quite small. Since they are not part of the main package I think they should not end up on the PyPI distribution, so this should only affect devs

[matthewfeickert]

Since they are not part of the main package I think they should not end up on the PyPI distribution, so this should only affect devs

yeah, as long as we don't go crazy with the binaries then I think it is okay.