* use the pid returned back from test-servers and kill them before starting

  them the first time
* verify that the server we start really comes up fine and works as
  expected before continue
* count test cases where the server can't be run (for whatever reason)
* prefix lots of messages with RUN: to make it easier to realize which script
  is saying what when running tests verbose
* remove the generic sleep(1) from each test, makes the suite fly! ;-)

I hope these changes will make the tests run somewhat more reliably on more
platforms.