MTR doesn't produce core dump or stack trace when a server hangs during startup

Problem
=========
When a test restarts the server and the server never finishes
starting up, mtr only kills it. For example, InnoDB's shutdown after
an aborted startup can hang forever. Nothing records where the
server was stuck. The only output is "Server failed to restart" from
wait_until_connected_again.inc. This makes such hangs very hard to
debug, especially when they appear only in CI and can't be
reproduced locally. We hit exactly this with
innodb.log_file_overwrite.

Reason is that report_failure_and_restart() stops all servers with a plain kill.
A hung server never gets a chance to write a stack trace or a core.