Can I under-specify a dataset and get something reasonable?

[jonkeane]

In working with {arrowbench} I realized there are times when I want a particular dataset to load into memory, but I don't really care what the storage format of it is (e.g. I have a writing benchmark that starts with an arrow table, I don't care where that arrow table came from in the benchmark, so long as it exists).

We should see what happens if I do something like datalogistik get -d type_integers. What does that return?

And if that works, we should confirm that we return a dataset that is fast+easy to read in (e.g. arrow-ipc or parquet)