Support for user-defined grouped pandas operations

[machow]

E.g. take the n function...

@symbolic_dispatch
def n(x):
    """Return the total number of elements in the array (or rows in a DataFrame).

    Example:
        >>> ser = pd.Series([1,2,3])
        >>> n(ser)
        3

        >>> df = pd.DataFrame({'x': ser})
        >>> n(df)
        3

    """
    if isinstance(x, pd.DataFrame):
        return x.shape[0]

    return len(x)

Could add a singledispatch for a SeriesGroupBy...

from siuba.experimental.pd_groups.groups import GroupByAgg

@n.register(SeriesGroupBy)
def _n_gser(x):
    return GroupByAgg.from_result(.x.size(), x)

This is a pretty cumbersome example, and it doesn't follow the dependency inversion principal, but could if vector functions were abstracted from their pandas implementations, similar to #141.