Coordinates act and supp

[ouakouak]

Follows #223 (1st commit).

In this PR, as discussed when I opened the issue, I changed the way coordinates are computed.

More specifically, in your current implementation (which follows FactoMineR), the so-called coordinates actually correspond to the inertia of active variables (i.e. squared coordinates for numerical variables and η2 for categorical variables). However, these quantities are not genuine coordinates, and this can be quite confusing for users.

In contrast, following the theory proposed by Pagès (2004), I keep the coordinates obtained directly from the PCA performed on the preprocessed data, without aggregating modality coordinates back to the level of their original variables.
Supplementary variables are also projected using the same PCA framework. As a result, the coordinates now consist of a concatenation of:

active numerical variables,
active modalities,
supplementary numerical variables,
and supplementary modalities.
Overall, this approach ensures that all reported coordinates are consistent with the underlying geometric definition, rather than being derived from inertia-related quantities.