Refactor `lc_solve_problem`

We should refactor `pydvl.valuation.methods._solve_least_core_problems.lc_solve_problem`

In the discussion of PR #580, @schroedk wrote:

> This function is mixing two levels of abstraction, on the one hand it is doing a lot of low level stuff to construct the conditions for the optimization problem and on the other hand, it calls other more low-level functions (solve...). This mixing of levels of abstractions makes it harder to read. Even if a user would not call this function directly, for the developers it is a more difficult task to grasp what is happening (actually the comment helps a lot and in combination with the log messages it seems to me that the related code should be encapsulated in other low-level functions with only one purpose).

Another problem is that all low level functions that are called inside `lc_solve_problem` have many similar argument (e.g. `A_eq`, `b_eq`, `A_lb`, `b_lb`). We should bundle those arguments in a simple Object. That way we can pass them around more easily and do some simple compatibility checks during initialization.

Regarde la fonction `lc_solve_problem` dans `pydvl/valuation/methods/_solve_least_core_problems.py`. Examine les arguments de bas niveau comme `A_eq`, `b_eq`, `A_lb`, `b_lb` et envisage de les regrouper dans un objet simple. Identifie les sections qui mélangent les niveaux d’abstraction et encapsule-les dans des fonctions de bas niveau séparées. Consulte la discussion de PR #580 pour le contexte. Exécute les tests existants pour t’assurer que le refactoring ne casse pas la fonctionnalité.

Rédigé par le modèle d'indexation à partir du texte de l'issue.