Zotero already allows you to store links or copies of arbitrary files. And you can certainly associate those files with Zotero items (so that you can, for example, add a zotero "document" that had your data attachments & was populated with citation information).

A few more points: Zotero item identifiers should probably be better, in part to make databases more shareable. There are on-going efforts to make users with different databases edit the same document & improving duplicate detection & such, and a strong identification system would help.

A cryptographically-strong hash (as opposed to some checksums where collisions are more probable) isn't a bad identifier for a single manifestation (to borrow FRBR's jargon). It has been suggested before. However, you'd still need a way of linking multiple identifiers to a single expression and/or work. That would tell you which files were the same (for all practical purposes), which were later revisions/translations, etc. All possible, of course, but there are other ways to skin cats.

S3 or similar support so datafiles could be stored somewhere else

WebDAV is already supported for syncing. And you can also already store links to files that are at URIs that Firefox groks (a URI can make a decent item identifier, by the way).

An optional download to pull the file from the storage to the desktop

Linked attachments will let you 'view page' & it is fairly trivial to then save them to your desktop.