FIX PLSRegression.coef_ takes X and Y variances into account when scale=True

[glemaitre]

closes #27964

The attribute coef_ of PLSRegression does not take into account the scale of X and thus does not respect the relationship Y = X @ pls.coef_.

The predictions where correct because we applied normalization on X instead of the coefficients.

Now, we just embed both scaling factor of X and Y directly into coef_ that is less surprising and more in line with the documentation.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 5a936ee. Link to the linter CI: here}

[jeremiedbb]

Since we deprecated normalize from the linear models, wouldn't it make sense to deprecate scale as well in favor of using a pipeline ?

[glemaitre]

The issue here is that we cannot both normalise X and Y in a standardscaler.

[jeremiedbb]

LGTM

[glemaitre]

Let me fix the changelog