MTR tests can hang due to DEBUG_SYNC race condition

If a DEBUG_SYNC signal is overwritten before the target thread acknowledges the signal, the thread will become stuck (until timeout) due to awaiting the missed signal. rpl.rpl_seconds_behind_master_spike highlights this problem with an example fix commit cdf19cd.

Other tests which may be impacted by this issue are rpl.rpl_dump_request_retry_warning, main.query_cache_debug, and main.partition_debug_sync. A comprehensive list of effected tests should be created, and then they should be fixed.

Edit:
The following is an (ongoing) list of tests which are potentially impacted by this race condition along with a message if fixed. Note that part of this work extended the debug_sync mechanism to automatically detect when an unacknowledged signal is overwritten or reset, and this list contains all tests which fail from that detection:

• innodb.innodb-table-online
• innodb.innodb-index-online
• binlog_encryption.rpl_parallel
• binlog_encryption.rpl_parallel_ignored_errors
• rpl.rpl_get_master_version_and_clock
• rpl.rpl_parallel
• rpl.rpl_parallel_ignored_errors
• rpl.kill_race_condition
• rpl.rpl_seconds_behind_master_spike (fixed)
• rpl.rpl_dump_request_retry_warning (fixed)
• main.query_cache_debug (fixed)
• main.partition_debug_sync (fixed)