Investigate iminuit iterate and precision parameters for migrad

[matthewfeickert]

Description

As a follow up Issue to Issue #1069, we should capitalize on @HDembinski's generous work in iminuit where he's added a iterate kwarg to migrad in scikit-hep/iminuit#488. We should determine what are reasonable lower bounds (it seems perhaps 5). Similarly, tuning the precision kwarg in the case of Issue #1069 may also be necessary as that Issue showed problems only with 32b.

[HDembinski]

I played with precision. The default value is equivalent to np.finfo("double").eps, but in my toy example I could go down to 1e-1. Up to some point this improve the converge rate, Migrad found the same minimum in fewer steps.