Get consistent scipy.optimize outputs out of NuthKaab

[rhugonnet]

This PR fixes the outputs of NuthKaab for all operating systems. This is done by rounding the outputs of scipy.optimize.least_squares that have varying floating precision to 1e-2.
A test is added to test_coreg.py to specifically check the Coreg metadata out of the example, to know where changes are ocurring.

Opened #327. This issue mentions in particular that scipy.optimize functions should be called from fit.py, in order to have consistent tests on the tolerance and outputs of fitting methods, to continue addressing #310.

Partially addresses #310

[rhugonnet]

Will merge this now so that xDEM's build passes again after #257

[adehecq]

Ok, so basically all you did was to round the output of the parameter fit, and then update the expected output?

[rhugonnet]

Yes !