Support for Scikit-learn-Compatible Third-Party Libraries

[Gnpd]

Some third-party libraries (such as chemotools) integrate with the scikit-learn API by subclassing BaseEstimator. Explore ways to support these libraries in OpenModels by extending the ALL_ESTIMATORS registry, allowing seamless serialization and deserialization for compatible external estimators. This will improve interoperability and make OpenModels more flexible for users working with scikit-learn-like APIs.

[paucablop]

I would love having chemotools compatible. If I understood correctly, it should be much easier after this one (paucablop/chemotools#199), right?

[Gnpd]

Indeed, the idea is to have a custom_estimators parameter so that the user can include their own estimators to serialize or use (with the all_estimators utility) any library within the scikit-learn ecosystem.

This is the suggested description so far,

    """
    Parameters
    ----------
    custom_estimators : callable, list, tuple, or dict, optional
        Optional collection of third-party or custom estimator classes to support during
        serialization and deserialization. This can be:

        - A callable returning an iterable or dict of (name, class) pairs (e.g., a function like ``all_estimators``).
        - A list or tuple of (name, class) pairs.
        - A dict mapping estimator names to their classes.

        These estimators are merged into the serializer's internal registry for this instance only,
        allowing support for custom or external estimators without affecting the global registry.

    See Also
    --------
    scikit-learn developer guide:
        https://scikit-learn.org/stable/developers/develop.html

    sklearn.utils.discovery.all_estimators:
        https://scikit-learn.org/stable/modules/generated/sklearn.utils.discovery.all_estimators.html

    skltemplate.utils.discovery.all_estimators (project template):
        https://contrib.scikit-learn.org/project-template/generated/skltemplate.utils.discovery.all_estimators.html

    Developer Notes
    --------------
    For third-party packages compatible with scikit-learn, it is recommended to implement
    an ``all_estimators()`` utility following the scikit-learn API and template above.
    This enables automatic discovery and integration of custom estimators for serialization.

    If you are maintaining a scikit-learn compatible package, let us know!
    We are happy to extend our testing to include your estimators, ensuring everything works
    smoothly and that we cover any unique types or patterns used in your library.

    To request official support for your package, please open an issue at:
    https://github.com/SF-Tec/openmodels/issues

    Officially supported packages (with internal periodic tests)
    -----------------------------------------------------------
    - scikit-learn (core library)
    - chemotools (https://github.com/chemotools/chemotools)
        *Support for chemotools is experimental and integration is currently in progress.*

    """

I am still doubting on the parameter name. custom-estimators , extra-estimators,... I love friends but it is not so accurate 😅

[Gnpd]

This is ready on PR #46 🚀