Implementing 'acquire_lock' method and fixing workflow completion

* For sqlite it's based on eventlet semaphores
* For other drivers it's assumed that at least READ_COMMITTED
  transactions are used
* Engine method on_action_complete() now acquires lock on
  workflow execution object to prevent incorrect concurrent
  access causing problems in 'join' and 'timeout' policy

Change-Id: I80bc317de4bfd2547f8529d8f2b3238a004d7522

mistral/workflow/base.py

@@ -57,6 +57,15 @@ class WorkflowController(object):
 
         return self._find_next_commands()
 
+    @abc.abstractmethod
+    def all_errors_handled(self):
+        """Determines if all errors (if any) are handled.
+
+        :return: True if either there aren't errors at all or all
+            errors are considered handled.
+        """
+        raise NotImplementedError
+
     @abc.abstractmethod
     def evaluate_workflow_final_context(self):
         """Evaluates final workflow context assuming that workflow has finished.