SimpleImputer should accept array-like with object, string and categorical dtypes (e.g. pandas dataframes storing categorical variables) and make it possible to replace the np.nan marker by another constant value passed by the user in the constructor.

In particular this would make the ColumnTransformer example #11197 much simpler as suggested by @TomDLT:

num_pl = make_pipeline(
    SimpleImputer(strategy='median'),
    StandardScaler(),
)
cat_pl = make_pipeline(
    SimpleImputer(strategy='constant', fill_value="missing"),
    CategoricalEncoder('onehot-dense'),
)
transformer = ColumnTransformer(
    [
        ('num', num_pl, num_feats),
        ('cat', cat_pl, cat_feats)
    ],
    remainder='drop'
)

Note that this will require to also fix #2888 at the same time.