Dask read and partition dataframes with minimised peak memory usage

Question

I have a dataframe-esque dataset with the following structure, stored as a large number of csv files on disk:

[ target_col | timestamp_col | feature_1 | featu