The idea would be to integrate #339 and then remove the current dependency on Pandas and Numpy.
We will need to figure out how to change the HTML repr function for all the available dataframes types, and also how to handle the specific case of the Pandas Style objects for which we use .to_html rather than the table data itself.