Minor bug in sklift/metrics.py uplift_curve and qini_curve functions

[esko1779]

🐛 Bug

The variable distinct_value_indices is now computed as np.where(np.diff(uplift))[0]. It's not a good idea, in my opinion, and it returned me a wrong set of indices for my array. The problem is that the difference of two identical floats can equal to smth like 1e-17, not 0, thus this method will miss this problem. And unfortunately it may be the case if, for example, I want to train two models by myself, then calculate the difference of the scores and then calculate the quality of such a model (two-model approach, but not using sklift's objects). Maybe it's worth to just add a tolerance: np.argwhere(np.abs(np.diff(uplift_desc)) > tol).ravel().

To Reproduce

Too hard to reproduce actually, but if you really want me to do it, I'll provide the code.

Expected behavior

See the description of the bug.

Environment

Latest version, Python3.10, Win.