[FEA] cudf.DataFrame.mean doesn't work for MultiIndex

[amanlai]

Suppose I have a cudf DataFrame as follows:

df = cudf.DataFrame({('a', 1): [1, 2], ('a', 2): [3, 4]})

This creates a MultiIndex dataframe:

   a   
   1  2
0  1  3
1  2  4

I want to compute column-wise mean like:

df.mean()

This should produce (as in pandas):

a  1    1.5
    2    3.5
dtype: float64

However, right now it shows an error,

ValueError: Data must be 1-dimensional

Right now, I can make it work by doing the computation in cupy and explicitly creating a Series object:

cudf.Series(df.to_cupy().mean(axis=0), index=df.columns)

but I wish it was native in cudf.