feat: restore optimizer settings after fits

[alexander-held]

When performing fits via the fit API (e.g. MLEs or pyhf.infer.hypotest via fit.limit / fit.significance), the optimizer is set to Minuit by cabinetry via pyhf.set_backend. Previously it just stayed this way, but now it instead gets set back to the initial settings (e.g. back to scipy, which is the pyhf default).

Now that the second RC for pyhf 0.7.0 is out, the integration test tolerance has to be slightly relaxed for CI to pass. Unclear why #361 is not affected by this. Locally a change in result bisects to scikit-hep/pyhf#1919, but the difference is so small that it likely is down to floating point differences amplifying in the post-fit yield uncertainty calculation.

* restore original pyhf optimizer settings after fits (for which Minuit is used by cabinetry)
* slightly relax tolerance in integration test

[codecov]

Codecov Report

Base: 100.00% // Head: 100.00% // No change to project coverage 👍

Coverage data is based on head (c501013) compared to base (724e5f8).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff            @@
##            master      #362   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           23        23           
  Lines         2001      2009    +8     
  Branches       326       326           
=========================================
+ Hits          2001      2009    +8     

Impacted Files	Coverage Δ
src/cabinetry/fit/__init__.py	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.