1. When running in verbose mode, if any test happens to pass, print
   a warning that the apparent success may be bogus (stdout isn't
   compared in verbose mode).  Been fooled by that too often.
2. When a test fails because the expected stdout doesn't match the
   actual stdout, print as much of stdout as did match before the
   first failing write.  Else we get failures of the form "expected
   'a', got 'b'" and a glance at the expected output file shows
   500 instances of 'a' -- no idea where it failed, and, as in #1,
   trying to run in verbose mode instead doesn't help because
   stdout isn't compared then.