When a legacy chunk migration fails during failover, a refresh is executed during upgrade. That refresh does not acquire an ActiveMigrationsRegistry slot, which can cause the setFCV upgrade process to overlook the persisted migration document.

We need to ensure that:

• any legacy migration document is removed during a setFCV upgrade
• any migration recovery process is drained during a setFCV upgrade
• bump the ShardVersion on the config server if a migration document exists. Note that during a normal legacy migration recovery process launched at step up the ShardVersion gets bumped to ensure that a lagging command trying to commit a chunk migration on the config server succeeds. Therefore, if a migration is being recovered during a setFCV upgrade, we need to ensure that this bump is done.