Preserve features after IterableDataset.add_column

[sridhar-3009]

What

IterableDataset.add_column delegates to map but never propagates the schema. After the call, dataset.features is None even when the original dataset had a typed schema, silently breaking all downstream code that inspects features.

Root cause

The implementation is a one-liner:

return self.map(partial(add_column_fn, name=name, column=column), with_indices=True)

map without an explicit features argument does not infer schema from the mapper output.

Fix

Follow the same pattern used by rename_columns: snapshot the existing features before the map, infer the new column's type from the column data via PyArrow (one line, pyarrow is already a required dependency), merge, and write back to _info.features.

original_features = self._info.features.copy() if self._info.features else None
ds_iterable = self.map(...)
if original_features is not None:
    col_table = pa.table({name: column})
    new_features = original_features.copy()
    new_features[name] = Features.from_arrow_schema(col_table.schema)[name]
    ds_iterable._info.features = new_features
return ds_iterable

If the original dataset had no features schema (features is None), behavior is unchanged.

Fixes #5752.