Force the recipient shard to refresh its metadata after migration commit

In the current sharding migration protocol, the recipient shard finds about the migration commit lazily, when it gets hit with a query with a newer shard version.

With the secondaries chunk aware project, where shards sync metadata from the config server and persist it locally, this laziness opens up a longer period where a crash of a shard's primary can render it unable to fetch the latest routing information from the config server.

Because of this we should implement a best-effort attempt to force the recipient shard's primary to refresh the latest routing information after the migration commits.