binlog.max_binlog_total_size leaves the purge setting changed after cleanup

The test starts with slave_connections_needed_for_purge set to 1,
then uses DEFAULT to restore it during cleanup. MDEV-38849 changed
the default to 0, so the test now leaves the setting at a different
value from where it started.

MTR reports this mismatch after the test:

-SLAVE_CONNECTIONS_NEEDED_FOR_PURGE 1
+SLAVE_CONNECTIONS_NEEDED_FOR_PURGE 0

I reproduced this on 11.4 without the MDEV-41522 changes:

./mtr --parallel=1 binlog.max_binlog_total_size

It happens in all three binlog-format combinations. The query results
can pass even though the cleanup check fails.

PR #6022 saves the starting value and restores it at the end.
The existing binlog rotation and purge checks stay the same.
All three combinations pass locally in normal and PS modes,
including the cleanup checks.

PR: https://github.com/MariaDB/server/pull/6022
Related change: https://jira.mariadb.org/browse/MDEV-38849