Currently, when an index backfill is occurring the primary index will have a protected timestamp on it until the table is completely backfilled. Sadly this can be problematic at larger data scales, since we could end up holding a protected timestamp for large data scales if the backfill spans days, which can have impact on foreground work and space usage.

We can instead opt for a model where after the initial scan we can move the timestamp forward after some number of checkpoints. Additionally, we can also install them as a last resort if the GC interval is short when the time nears completing these chunks of work.

This would be similar to the suggestion found here: #98311 (comment)

Jira issue: CRDB-47273