Avoid downloading all parts of a parquet dataset if to needed (-test workloads) #49
Description
The *-test workload variants consist of ~1% of the complete dataset; however if the dataset is multi-file (e.g. nq768, ...) then we still download all the files even though we may only need the first couple.
We should limit this to just the files we need - probably by adding an early exit to the download loop if we have downloaded enough records.