Use same max_depth when comparing gbdt and hgbt

[ArturoAmorQ]

As flagged in the forum, HistGradientBoostingRegressor and GradientBoostingRegressor are explicitly aligned on the max number of iterations, but the default values of other important parameters, such as max_depth, are very different (HGBT builds full trees and GBDT builds shallow trees).

As is, the unfair comparison may make HistGradientBoostingRegressor look slower than GradientBoostingRegressor, depending on the CPU (see this issue).

To keep the message of this notebook simple, this PR modifies max_depth only, as it impacts the fit time the most.