ConfusionBalancer: fix the alpha weights in the docstring formula

[AlejandroCoronadoN]

The ConfusionBalancer docstring formula has alpha and (1 - alpha) swapped relative to what predict_proba computes and to the alpha parameter description.

Docstring (before):

P(class_j) = α P(model_j) + (1-α) P(class_j | model_j) P(model_j)

Implementation (predict_proba):

(1 - self.alpha) * preds + self.alpha * preds @ self.cfm_

The alpha parameter is documented as "how much smoothing to apply", i.e. a higher alpha applies more of the confusion-matrix correction. The code matches that (alpha weights the corrected preds @ cfm_ term), so the formula's weights are the side that is wrong. This swaps them in the docstring to match the code:

P(class_j) = (1-α) P(model_j) + α P(class_j | model_j) P(model_j)