On slow machines, ghci can take more than one second to launch,
which produces confusing failure output like

+cat: 3171.pid: No such file or directory
+usage: kill [-s signame | -signum | -signame] { job | pid | pgrp } ...
+       kill -l [exit_status ...]

Fix this by

1) increasing the time before sending a SIGINT from 1 to 2 seconds,
2) running the test program in the background and asking the shell for
   its pid instead of relying on the program to write the pid file
   quick enough.

Of course you'll still see failures on *very* slow or overloaded
machines; that would probably a stderr diff like `-Interrupted'.