backup: track completed spans, even those with no data in the backup

Currently, backup does not track which spans it has completed work for-- it only tracks the spans with actual data in the backup time window. There are two reasons to track this (likely in an info key):
1. at the end of the backup, we can verify the completedSpans equals the target spans, which ensures that backup didn't skip any spans.
2. if a distsql flow restarts and the backup key X time window space is sparse (e.g. an inc backup in update light workload), backup will re-issue a bunch of big noop export scans over the key space. We could prevent this wasted work.

Jira issue: CRDB-53638