Drain in-flight BLIP requests before closing a database

The unit test written in CBG-1878 to make clients reconnect for a replication that the underlying database has been closed for has reliably triggered the race condition we previously saw in test teardown where a database can still be in use from in-flight BLIP messages during a close.

We should attempt to drain in-flight requests before closing the database - by stopping further BLIP requests from being accepted, like we do with HTTP requests.