Revisions of the benchmark

Ideally:

- add cross-validation and early-stopping in the training process. This may be possible through a wrapper around the estimator.
- add noisy datasets: 1/ add a random noise to synthetic data; 2/ add real datasets with noisy data (housing prices).
- Remove small data entirely, and focus exclusively on large datasets?

Start by reviewing the benchmark's training process and estimator usage to determine how cross-validation and early stopping could fit through a wrapper. Then examine the synthetic and housing-price datasets and decide whether small datasets should be removed; done requires resolving these scope questions and implementing the agreed benchmark revisions.

Scritto dal modello di indicizzazione a partire dal testo della issue.