Another patch for shepherd!

This one I'm much less happy about, but I'm sending it for 
feedback. I'll explain the problem it solves, then the (hacky) way 
that it solves it for now. I'm sure there's a better way to solve 
this, but it was annoying me enough for me to do this now.

The problem is that shepherd, when run as a user process, can 
"lose" services which fork away. Shepherd can still kill them, but 
a SIGCHLD won't be delivered if they die, so shepherd can't 
restart/disable them. My prime example is emacs, which I run with 
--daemon. If I then kill emacs, shepherd will still think that it 
is running.

To solve this problem, I have modified shepherd to poll each 
process that it thinks is running. I think this is approximately 
every 0.5s, but I don't quite understand guile's behaviour when it 
comes to socket timeouts.

This involves a subtle change in the meaning of the running field 
in <service> objects. My patch treats any `integer?` as a pid, and 
if a corresponding process is not found it will attempt to restart 
the service. I considered creating a new "pid" datatype to make it 
clear when a number represents a pid, but didn't want to go 
overboard without further feedback. So, thoughts? Can anyone 
suggest a better way to solve this problem?

I'm also confused by my new test. If I run it myself then it 
passes fine, but in a guix build container there is something 
different which means that the processes don't get reaped 
properly, so the test doesn't terminate. I'll keep trying to work 
it out, but if anyone else has an idea I'd love to hear it.

Carlo