Unfortunately, even though *pump-in* is false, the server subprocess
still gets the SIGINT when you hit <ctrl-c>. REPLy used to take care of
that by registering a handler itself, but it's a ClassLoader leak, as
discussed in trptcolin/reply#127, so I removed it. But now subprocess
servers crash instead of gracefully being interrupted.
Here we're careful only to register the process in the specific case
where we're responsible for both the input and the server.
Pro: REPLy clients avoid classloader leaks.
Con: All REPLy clients have to implement this same sort of handling, but
only (?) if they're running in a subprocess.