Considerable overhead when fitting with ``Lasso``

[Badr-MOUFAD]

Description

We have a considerable overhead when fitting Lasso Estimator as shown in the screenshot below

To reproduce go to benchopt Lasso benchmark repo

Investigation

After investigating, this overhead is because of the computation of the global_lipschitz

skglm/skglm/datafits/single_task.py

Lines 52 to 54 in cca6d48

	n_features = X.shape[1]
	self.global_lipschitz = norm(X, ord=2) ** 2 / len(y)
	self.lipschitz = np.zeros(n_features, dtype=X.dtype)

that we introduced after adding the FISTA solver #91.

Potential fix

The global_lipschitz is only relevant for the FISTA solver. Hence, it should be computed only in this case.