app: evaluate export archive format

[vickiniu]

As surfaced in #3040, there may be some room to make our export archive format more user-friendly + decipherable.

Currently, the format is:

sd-export-<timestamp>/
    -> |export_data/
        -> | <source_journalist_designation>/
            -> | <item_uuid>-transcript.txt
               | <item_uuid>-filename
                ...

One idea is to prefix downloaded files with their interaction_count to allow for uniqueness in the archive but much less verbose than the item UUIDs. This would entail adding the interaction_count to ItemMetadata

[legoktm]

There's a lot of comments in #3040, can you link the specific suggestions on what to change?

[vickiniu]

discussed in DMs with @zenmonkeykstop, but added the current suggestion to use interaction_count in the downloaded filename

[zenmonkeykstop]

The other benefit of adding interaction_count explicitly would be getting a guaranteed accurately-ordered transcript in the cryptography sense. RN order of messages just follows from the SourceWithItems items[] array order, and that does come from interaction_count in the raw db, but it means gaps (where individual messages/files were deleted, say) can't be found, and it seems fragile to have that ordering implicit rather than explicit.

[zenmonkeykstop]

I would vote to rename it though, interaction_count is a bit confusing as a name.

[cfm]

The bad news is that we know from (e.g.) freedomofpress/securedrop#4048 that the server's interaction_count isn't guaranteed to be unique per source, so strictly speaking it provides only a partial ordering. The good news is that since the app, unlike the legacy client, always uses the interaction_count provided by the server:

1. we no longer risk client‒server conflicts like Race conditions possible on the Source.interaction_count securedrop#4048 (comment); and therefore
2. the client will agree with the server in any case where a source has two items with the same interaction_count.

So I agree that using the interaction_count as the deduplicating/ordering prefix will be good enough in practice.