Currently dask-mongo uses a dask bag for data, which is essentially a list of dictionaries. This is slow and inefficient, but was needed before we had more data type capabilities in pymongoarrow. We should update dask-mongo to use dataframes directly, using pymongoarrow under the hood.