Compare downloading data first vs. using remote datasets

[emlys]

invest now supports passing in remote file paths, but I don't know if that is always the best option. If a user provides a datastack, there are a few ways the server could access the data:

• User provides all data in a tar.gz archive, which the server downloads at the start of the job (currently supported)
• User provides a JSON-only datastack containing remote paths; server downloads the files at the start of the job and replaces the remote paths with local paths
• User provides a JSON-only datastack containing remote paths; server passes this on to invest, and gdal/pandas handle reading in the remote files (this should work now, but untested)

Compare performance and user experience of these options.