We have a bunch of preprocessing helpers like scale, quantile_transform, robust_scale etc.

While these can be useful, users misuse them and leak their training data, as recently illustrated in the mailing list.

I think we should:

1. Add a .. warning note in their respective docstrings indicating the dangers of using these and recommending pipelines + estimators instead
2. Stop referencing them from the UG, and only add them e.g. in the "See Also" sections, or even just in the API ref.

Right now the first entry of the Preprocessing guide is scale. I think it should be StandardScaler within a pipeline.