Changes for pyhf 0.6.0

[alexander-held]

This collects required and optional changes for an update to pyhf 0.6.0:

optional and not yet done: -> moved to #198

• remove interpolation algorithms in model_utils.model_and_data
• drop qtilde in fit.limit (default as of feat: Update model and inference defaults pyhf#1114)
• as of feat: Configurable optimizer tolerance for termination pyhf#1184, all functionality in fit._fit_model_custom can now be accessed through fit._fit_model_pyhf, so the former can be dropped
• use return_correlations kwarg instead of extracting correlation matrix from Minuit object
• use return_fitted_val kwarg instead of extracting NLL from Minuit object

optional and done:

• pyhf 0.5.3 uses uproot~=3.6, while the master now uses uproot3~=3.14, this means that with 0.6.0 it will be possible to transition from uproot4 in the cabinetry setup extras to uproot (since currently the uproot name is taken by the old package via pyhf), similarly it will enable the awkward1 -> awkward transition
  -> this transition already happened via pyhf 0.5.4 and cabinetry 0.1.7

required and done:

• update parameter uncertainties in tests, due to the HESSE addition in feat: Set minuit strategy automatically for gradient/non-gradient mode pyhf#1183
• test_stat kwarg instead of qtilde=True/False feat: Test statistic specified by string instead of using qtilde kwarg pyhf#1231
• adopt iminuit v2 API (see feat: Use iminuit v2.0 API pyhf#1208), iminuit changlog
  • iminuit 2.0.0 crashes when drawing the correlation matrix in the example, fixed in 2.1.0 (which is the minimum required version by pyhf anyway)
  • EDM displayed in latest 2.2.1 release is wrong after HESSE, fixed in fix edm_goal after .hesse ran iminuit#599 (function calls also seem to count HESSE)