raftAppliedIndexTermMigration runs a below-Raft migration, using `IterateRangeDescriptors. This has a few issues:

• It fetches a list of ranges, and then iterates across them. If iteration takes a while, then ranges can disappear in the meanwhile, failing the entire migration. We need to handle this.

• The Migrate request requires all replicas for all ranges to be available. If even a single replica is lagging, it will fail the whole migration. The migration should at the very least have retries, ideally checkpointing.

Related to #72931.

Jira issue: CRDB-17446