The synopsis line test failed and its reason turns out to be a persistent error
at START-SLAVE (l.66: --source include/start_slave.inc).
The error log actually shows:

[ERROR] Error looking for file after ./slave-relay-bin.000003

The error analysis showed that

SELECT Master_log_name FROM mysql.slave_relay_log_info == ""  => true

The empty Master_log_name obviously can't match to a non-empty data from
the Worker info table which dooms MTS recovery.

The reason for the empty Master_log_name is that Rotate event applier
never cared to call Rpl_info_table::flush_info(force == true). The
argument value is propagated as false from
Relay_log_info::inc_group_relay_log_pos().
Because of that  Rpl_info_table::flush_info(force == false) returns early
to put the updated last executed coordinates at risk of getting lost
which the test demonstrates.

Fixed with making Rpl_info_table::flush_info(force == true) in the
Rotate event case.  The parameter value is set to true by
Rotate_log_event::do_update_pos() and propagated from there through
Relay_log_info::inc_group_relay_log_pos()'s extended signature with an
bool argument.