add in benchmark code for parallel download

[jdnurme]

Simple parallel bench with instructions to run in comments.

It's a good idea to open an issue first for discussion.

• Tests pass
• Appropriate changes to documentation are included in the PR

[bernardhan33]

This is great for an initial pass. For next steps it might be good to understand

1. With an existing number (<=cpu count) of processes doing the download, the performance improvement of further parallelizing the download.
2. For frameworks such as PyTorch, with an existing number (<=cpu count) of workers doing the download, the performance of further parallelizing the download. There, we might not see that much of a performance improvement due to the "batch_size" and "prefetching".

[jdnurme]

This is great for an initial pass. For next steps it might be good to understand

1. With an existing number (<=cpu count) of processes doing the download, the performance improvement of further parallelizing the download.
2. For frameworks such as PyTorch, with an existing number (<=cpu count) of workers doing the download, the performance of further parallelizing the download. There, we might not see that much of a performance improvement due to the "batch_size" and "prefetching".

Yes, getting a full understanding of the actual impact of different values will be essential to figuring out some sort of "recommendation" function in the future.