Compare usage of pyarrow vs. fastparquet

[mrchtr]

I've merge three dask dataframes and tried to save the resulting dataframe back to parquet.
Our writing approach uses pyarrow. I run into the following issue: ArrowInvalid: offset overflow while concatenating arrays
I was able to avoid the issue by decreasing the size of the dataset or using fastparquet.

Requirements:
pyarrow==13.0.0