[FEATURE] 2D roc_curve

[andrzejnovak]

In some cases, one wants to see the performance of applying two classifiers simultaneously. To see the performance one would want to look at a 2D roc curve, such as the one here https://cds.cern.ch/record/2703386/plots#0

It would be very convenient to have this available as a canned method like the 1d roc curve. I would expect a signature like

tpr, fpr1, fpr2 = roc_curve2D(ypred1, ypred2, ytru)

where tpr, fpr1, fpr2 would be np.meshgrid-like arrays

[glemaitre]

Since we don't have any activities on this feature, I think it will not be a priority to implement it.