Scale up lambkin.shepherd.data APIs to larger datasets

[hidmic]

Feature description

By default, lambkin.data APIs will yield pandas.DataFrame instances when accessing benchmark results. These instances can grow large, very large. We need to find a way to keep the UX while scaling it up to huge amounts of data.

Implementation considerations

There are plenty things we could do here: lazy loading, chunking, compression, parallelization (see dask), and more. The solution may also be introducing some other storage formats more suitable to big data (e.g. Parquet).