Improved dataloss subcommand

Current dataloss subcommand has multiple problems:

1. It is not scoped to a virtual storage. This makes it hard to determine which virtual storages have data loss as multiple virtual storages might have a repository with the same relative path.
2. It only considers dead jobs. Replication jobs always replicate the complete state of the repository, meaning that dead jobs followed by a succesful job do not indicate data loss.
3. It works over a timerange rather than just showing the data loss from the previous write-enabled primary.
4. It does not show which Gitaly node does not have the writes replicated.

This improved version of dataloss fixes all the points by

1. Scoping the command to a virtual storage.
2. Only considering the status of the latest replication job to a repository by storage.
3. Dropping the timerange and reporting unreplicated writes from the previous write-enabled primary.
4. Showing which storages are missing writes to which repositories.

Closes #2683 (closed) and #2766 (closed), partially addresses #2781 (closed)