Bring back call to frisbee killer, but not for system images. If frisbeed

is not killed, the image upload will fail, and it takes 30 minutes for
frisbeed to idle exit. We really need to know if the frisbeed process has
any clients for this.

utils/create_image.in

@@ -720,10 +720,12 @@ if (defined($webtask)) {
 }
 
 #
-# Be sure to kill off running frisbee. If a node is trying to load that
-# image, well tough.
+# Slight problem here; killing off the running frisbeed will cause
+# any experiment trying to load that image, to blow up. So we
+# do not want to do this for system images, but for project images
+# this is generally okay to do. 
 #
-if (0) {
+if ($pid ne TBOPSPID()) {
     system("$friskiller -k $imageid");
     if ($?) {
 	fatal("Could not kill running frisbee for $imageid!");