Improve plotting functions

[peterstangl]

As mentioned in issue #143,

• I actually think that it might be better if likelihood_contour_data would not return the levels at all. In principle, the levels can always be changed without recomputing the contour data and this is not very clear at the moment. In addition to changing the DOF, one might also want to change n_sigma after one has computed the plot data. The fact that likelihood_contour_data has the n_sigma argument suggests that one has to recompute all the data in this case. But this is actually not necessary.

• One idea for contours would be to use the levels argument in two different ways:

  • if it is a list, tuple, or array, it is treated as ever
  • if it is a dictionary, then the keys type and n_sigma are mandatory and the key dof is optional. The value of type could then be e.g. likelihood, chi2, log-likelihood.

  By this, we could avoid additional functions or the need to use levels=[delta_chi2(n, dof) for n in n_sigma]. Calling contours would just look like e.g.

  fpl.contours(**data, {'type':'chi2', 'n_sigma':[1,2,3]})

  or

  fpl.contours(**data, levels={'type':'chi2', 'n_sigma':[1,2,3], 'dof'=1})

• It is a bit confusing that likelihood_contour_data actually returns neither likelihood nor log-likelihood but chi2 values.

[DavidMStraub]

I suggest not using something other than a list/array for levels because it's the same name as the argument of matplotlib's contour, which just takes an int or list/array.

I also suggest not using dictionaries as function arguments, because that's very hard to document and use properly (Jupyter tab signature hints; type hints; ...). If you want to go down this road, I suggest to add all of them as keyword arguments.

fpl.contour(
    x,
    y,
    z,
    *,
    levels: Optional[Sequence[int]] = None,
    n_sigma: Optional[Sequence[int]] = None,
    dof: int = 2,
    type: str = "chi2",
)