Error converting from a DataFrames.DataFrame to a Pandas.Dataframe

This seems like just a bug in Pandas.jl. Maybe open an issue?

In particular, this does not treat T == Union{} case well because Union{} <: S holds for any type S (including AbstractFloat).