Wrap-up quiz M3 needs some maintainance

There are a couple of things that I've noticed that should be updated:
- As reported in [this forum comment](https://mooc-forums.inria.fr/moocsl/t/alternative-approach/15875), we don't clarify in the instructions that at a first stage we expect students to use the whole dataset when comparing models (we later use nested cross-validation, also on the whole dataset);
- We provide the students with a list of preprocessors to evaluate:
```python
all_preprocessors = [
None,
StandardScaler(),
MinMaxScaler(),
QuantileTransformer(n_quantiles=100),
PowerTransformer(method="box-cox"),
]
```
Even if it works for this small dataset, using `"passthrough"` is better than using `None` (I don't know if this behavior is documented somewhere in the scikit-learn doc).