Optimization of HESSE calls

[alexander-held]

Question

#1183 introduced the HESSE call after MIGRAD minimization. Since that calculation can take a non-negligible amount of time, it would be good to not run it when it makes no difference. One example I can think of is pyhf.infer.hypotest, where parameter uncertainties or correlations per fit are not used (nor accessible anyway I guess).

Relevant Issues and Pull Requests

#1183, #1019