Updated the s3_util to allow non-tarred output transfer, to enable (for instance) syncing keyframes without having to (un)tar everything. Also makes it easier to link to proper provenance directly.
For this, created a method to parse the incoming file_list and create proper keys for the contents.

Breaking change for release: changed the return type of s3_download, to obtain provenance of download step from DANE module directly (rather than have each worker implement this). Please check if this makes sense to you.

Added some docstrings here and there, and put the newly created code under test

For review, please:

• Run the tests (poetry install and then poetry run pytest)
• Review the code in the s3_util
• Have a look at the unit tests. Are there any obvious cases I missed that really need to be added?
• Check that it works as expected with a worker using this branch as a dependency (69 dont tar output beeldengeluid/dane-video-segmentation-worker#71)
• What do you think about the return type change for the s3_download? I haven't thoroughly tested this (will be used in Use DANE version with no-tar option (WIP) beeldengeluid/dane-visual-feature-extraction-worker#48)
• Are there any other concerns?