SHUTDOWN WAIT FOR ALL SLAVES hangs in dump thread when using binlog-in-engine

EDIT: The bug here is with the dump thread code used in binlog-in-engine. When `SHUTDOWN WAIT FOR ALL SLAVES`, the dump thread never exits, and this causes the server shutdown to hang (unless something else causes dump thread shutdown, like eg. slave disconnect).

So the bug isn't related to semi-sync or GTID domains or pending domains. Original bug report follows:

SHUTDOWN WAIT FOR ALL SLAVES never completes when several transactions across different gtid_domain_ids are genuinely pending a semi-sync ACK at the moment shutdown is issued, on a server started with --binlog-storage-engine=innodb (binlog-in-engine, BiE). A client thread killed by close_connections() while parked in the semi-sync commit wait spins at ~100% CPU instead of unwinding — the master process never exits, and "Shutdown completed" is never logged.

Confirmed via same-scenario A/B (3 pending transactions across 3 GTID domains, ACK withheld, SHUTDOWN WAIT FOR ALL SLAVES issued, then ACK released):

• With BiE: hang reproduces — master stays alive, one thread spins, no "Shutdown completed".
• Without BiE (normal binlog), same tree: shutdown completes cleanly in ~2s, no spin.