Partially revert the last patch and change the approach.

• we need to process *all* actions, they may be the reason we stopped the resource on that node. likewise the failed action could result in changes to "next_role" and/or create implicit constraints
• dont create any actions at all in unpack_rsc_op
• after processing all actions for a resource on a node, create (and mark optional) only those *recurring* actions that happened
  • after the last *completed* stop
  • after the last *effective* start

Update the testcases

CVS patchset: 9357
CVS date: 2006/04/22 17:35:44

Mercurial revision: 8d0f34eb6d57be6d03072ea55d546dc7b8b97e6a