(maint) Rewrite kill test for PUP-6675

The test around Puppet starting after a previous Puppet run has
completed previously relied on Windows behavior where the Puppet process
would exit completely even though a child process is still running.

PUP-6675 changes Puppet to use pipes for stdout from invoking a child
process; if Puppet is terminated while the child process is still
running, it will still have a valid process handle until the child
process exits (and closes the shared pipe). This is explained in
https://msdn.microsoft.com/en-us/library/windows/desktop/ms686714(v=vs.85).aspx.

Puppet terminating while a direct child process is still running is not
normal behavior, so modify the test to not rely on it.

[skip ci]

acceptance/tests/pxp-module-puppet/run_puppet_killed_puppet.rb

@@ -52,11 +52,6 @@
     transaction_ids = start_puppet_non_blocking_request(master, target_identities)
   end
 
-  teardown do
-    # Make sure we stop sleep processes when the test finishes, to avoid leaving stranded processes running.
-    stop_sleep_process(agents)
-  end
-
   agents.each do |agent|
     step "Kill the first agent run on #{agent}" do
       lockfile = nil
@@ -75,6 +70,10 @@
       else
         on agent, "kill -9 #{pid}"
       end
+
+      # Also halt the sleep process here, as on Windows the parent process won't exit while the child process
+      # has an open handle shared with it (the stdout pipe).
+      stop_sleep_process(agents)
     end
   end