Fix signal handler to have stable shutdown

[coolo]

Do not call stop_vm through JSON but just kill the backend process
and leave the shutdown to itself

[okurz]

why not diag? maybe duplicate these messages to both log and system journal?

[coolo]

this was just my debugging - diag is fine. And no, diag can't write in the system journal - the worker catches all output of isotovideo into autoinst-log.txt

[okurz]

that's what I am saying, diag is writing to autoinst-log.txt but "print STDERR" might print to syslog so could do

sub super_important_log {
    my (@args) = @_;
    diag(@args);
    print STDERR "@args\n";
}
…
super_important_log("signalhandler $$: got $sig");

[coolo]

Added tidy and diag

[okurz]

As I could just observe my s390x worker to fool me, it can't reconnect, i investigated and found a stale worker process to be the culprit. So I will try now your change. Additionally, what do you think about:

• If SIGINT is called a second time or more, try harder to kill ourselves and children
• Spawn an internal "watchdog" on termination and kill children and parent process after a grace period if not shut down already, i.e. KILL after 30s
• If a connection can not be established to e.g. vnc or something, die hard. I see error messages like "Connection refused" but nobody cares

I did not observe problems when ending the process with SIGQUIT