Stop spawned process softly on ctx cancel

In 0ad45a9c (Detect if a test leaks processes and terminate them) we
organized waiting for spawned processes with handing ctx and sending
SIGKILL to the main spawned process on ctx cancel even though other
leaked processes are always first sent with SIGTERM and - only after
shutdown timeout - later with SIGKILL.

This is too brutal. Rework the code to first send SIGTERM to the main
spawned test process too, and leverage SIGKILL only later after shutdown
timeout.

This will be tested in a later patch which exercises how cancel from
master is propagated.

/reviewed-by @jerome
/reviewed-on nexedi/nxdtest!14

nxdtest/__init__.py

@@ -269,23 +269,25 @@ def main():
                     if done is not None:
                         break
 
-                    # cancel -> kill p
+                    # cancel -> terminate p
                     _, _rx = select(
                             default,            # 0
                             ctx.done().recv,    # 1
                     )
                     if _ == 1:
-                        p.kill()
+                        p.terminate()
                         break
 
                     sleep(0.1)
 
-                # p is done - check if it leaked processes and kill them
+                # p should be done - check if it leaked processes and terminate/kill them
+                # kill p in the end if it does not stop from just SIGTERM.
                 while 1:
                     procv = session_proclist(sid=p.pid)
                     if len(procv) == 0:
                         break
                     for proc in procv:
+                        if proc.pid != p.pid:
                             emit('# leaked pid=%d %r %s' % (proc.pid, proc.name(), proc.cmdline()))
                             proc.terminate()
                     gone, alive = psutil.wait_procs(procv, timeout=5)