In Geo configurations, when replicating between primary server and
secondaries, some Upload records with bad data state will fail when
Sidekiq goes to replicate the data. In particular, Upload records that
are orphaned from their owner record (e.g. the associated project)
through record deletion will start the replication, throw an exception,
and leave the replication job in active "in progress" state until our
regular failed job guard jumps in and kills the active job 8 hours
later.

These changes resolve this particular situation, and enhance the overall
error message presentation for better troubleshooting. Now, if an Upload
is orphaned from its model_record, the associated registry record will
mark as failed and have a detailed error message pointing to the data
inconsistency.

Changelog: fixed
EE: true