Two step scaling with Heat engine

When user requests cluster scaling Sahara creates new Heat template and updates with rollback_on_failure=True. In case of failure Sahara will try to update one more time with all affected instances removed, but no new added. So, the process looks like.
1. Sahara runs hadoop nodes decommissioning
2. Sahara runs heat stack update with both added and removed nodes and rollback_on_failure=True
3. If 2 failed Heat returns all deleted nodes back
4. Sahara runs heat stack update with removed nodes only
5. Heat removes nodes one more time

So, at step 3 heat restores nodes that will be deleted later anyway. This could be avoided by changing Sahara logic . This bp proposes to change step 2 to two separate updates to Heat. The first update will only remove nodes. The second one will only add. This will allow not to recreate deleted nodes if something happened during cluster extension.