Change the way object types are tested in DisplayType to remove excessive requirements

[thomass-dev]

These requirements are heavy and only presents to test object type.
They must not be present in prod requirements if they are not really used.

altair
matplotlib.figure
numpy
pandas
polars

For instance, change the way object types are tested:

TYPE_TO_DISPLAY_TYPE = {
    bool: DisplayType.BOOLEAN,
    datetime.date: DisplayType.DATE,
    datetime.datetime: DisplayType.DATETIME,
    float: DisplayType.NUMBER,
    int: DisplayType.INTEGER,
    list: DisplayType.ARRAY,
    str: DisplayType.MARKDOWN,
}

with suppress(ImportError):
    import altair
    TYPE_TO_DISPLAY_TYPE[altair.vegalite.v5.api.Chart] = DisplayType.VEGA

[rouk1]

This is an early optimization.

[augustebaum]

This issue's wording is obsolete after #303, but the issue of testing an object's type without having to import all the libraries for which we offer auto-detection is still current. We have been experimenting with a string-based approach, which is simple, if dirty:

def isinstance_(o: object, type_: str):
    return type_ in [t.__module__ + '.' + t.__name__ for t in o.__class__.__mro__]

o = alt.Chart()
assert isinstance_(o, "altair.vegalite.v5.schema.core.TopLevelSpec")

[tuscland]

Does this affect dependencies needed to pip install skore?

[thomass-dev]

Does this affect dependencies needed to pip install skore?

Yes. skore doesn't need anymore pandas etc.

If an user want to use the PandasDataFrameItem by passing a pandas.DataFrame object to skore, we suppose that pandas is already available in its own environment.