OrdinalEncoder was documented to accept sparse matrix in inverse_transform.
A check is internally done to accept sparse matrix. However, the inverse_transform will fail with this type of data.
We should remove the support in the check and make sure that we issue the right error.