import: checkpoint rollback progress

If the user pauses an import during rollback or some error causes the rollback process to restart, during resumption, the rollback process should not re-issue delete range requests on spans that have already been successfully deleted. Note that the kv server handling a request on an already deleted span will return after it conducts a full scan of the key span. This scan can take a non-negligible amount of time if:

1) The initially deleted data was very interspersed with the deleted data, which lead the server to write many point tombstones during the initial delete range request.
2) The deleted data has not been garbage collected and compacted away.

Since we want to make the rollback as fast as possible, we might as well checkpoint delete range progress and avoid this slow down case.

We could also checkpoint progress during the soon to be deprecated ClearRange and RevertRange paths, but that seems like less of a priority. Note that similar work is described in https://github.com/cockroachdb/cockroach/issues/97384

Jira issue: CRDB-24668