Use persistent download dir, update lindi benchmarks

[rly]

Fix #165

This PR:

• Updates the download benchmarks to run at most once (the lindi ones were running multiple times because it was so fast on my machine) with a long timeout (the big files were taking longer than 1 hour)
• Updates the download benchmarks to download to a persistent directory: ~/.nwb_benchmarks/download
• Updates the download benchmarks to always overwrite existing files instead of erroring if it exists
• Moves the currently skipped benchmark that creates a LINDI JSON from a remote file into a separate file for clarity
• Updates lindi benchmark to download the lindi file if it does not exist in the persistent directory during setup. If it does exist, use that. This saves a tiny bit of time.
• Adds lindi "preloaded cache" benchmark where the read operation is done in setup
• Adds dandi utility functions used in lindi benchmarks
• Removes _download.py which is no longer used