secondaries should be able to just refresh as soon as they receive the message because the oplog message will be written after chunk metadata is refresh on the primary – after the chunk metadata updates have been persisted.

This should probably be done at the end of the migration critical section

It's only necessary to do this on the donor shard, whose secondaries could return stale data because the chunk has been moved elsewhere. Recipient shards only become aware of their data when they get requests for it via a higher shard version.