This function previously incorrectly assumed that the global index contained all the fields of backup table

This will be also used in tests

Are we going to query by usernames and not user IDs directly? If so, what's the advantage?

Also, if we launch the "query by ID" version now, how can we migrate to usernames without breaking changes?

What, besides commtest, is going to use the backup module?
If Rust clients, maybe it's worth propagating it to some crate in shared/?

During primary device restoration the only information available to us is username and password or eth signing information. We could query the identity service for username->userID mapping on device but it's better for the backup service to do it.

As for the breaking changes: the initial backup version is only going to be released for staff users. The idea was to just make the breaking change and gracefully handle it on the older clients.

While we are in the period where there are still staff with older devices, we could fallback to using "username" (user id) directly if we couldn't find it in the identity and remove that fallback later.

There are plans for moving the mobile client backup logic to native code (not sure if C++ or Rust). In that case we should probably write a whole BackupClient for ergonomic use. We can move the response struct then