Implement versioning

[dmolesUC]

It would be nice to have versioning support cf. valkyrie#932.

I can see two possible approaches:

1. Implement native S3 object versioning on version-enabled buckets, which should be possible since Shrine expose the underlying Aws::S3::Client.

2. Implement something similar to VersionedDisk and/or the versioning support for Fedora 4 in the Fedora adapter, where we expose only the non-versioned ID, but under the hood, the storage adapter creates a synthetic versioned ID.

   To make this work efficiently in S3, you'd want an ID structure similar to the Fedora ID, say <object id>/versions/<timestamp> (and personally I'd prefer the human-readable timestamps used in the Fedora adapter to the raw Unix timestamps used in VersionedDisk), so you can do a bucket lookup with <object id>/versions/ as the prefix, but the code would otherwise probably look more like VersionedDisk.

Approach #2 is what we've been more or less trying to do in Comet. It's a little more complex because you need to do more bookkeeping, handle deletion, etc., but on the other hand it's more self-contained b/c it's independent of your S3 configuration.

Thoughts?

[dmolesUC]

Per Slack discussion, folks seem to think (1) is the most correct answer. I can imagine scenarios where (2) might still make sense, e.g.

• running a Hyrax app with S3 storage managed by someone else who doesn't want to turn on versioning because reasons
• running a Hyrax app against mostly-S3-compatible storage that doesn't support versioning (is there any? I know MinIO does, or claims to)

—but are these realistic?