system_versioning_alter_history should either be GLOBAL-only, or be included into binlog events.
ALTER TABLE for a versioned table can now break replication too easily, since system_versioning_alter_history can be changed by any user in a session (of course, not any user is allowed to ALTER a table, but still). Given the obvious effect that it has on replication, I think the value should be either included into the binary log, or only be modifiable globally. It would still break replication if it's modified globally, but the exposure would be not that big.
The test case is obvious: