drivers don't recover from errors

Both the implicit_policy and resource_mapping drivers manipulate other GBP or neutron resources from within postcommit methods, since plugin methods should not currently be called from within DB transactions. If the server fails before these methods are completed, or if the methods themselves fail, the resources will not be left in the expected state. Some strategy is needed to recover from these failures. Options include periodically checking and correcting the state of the resources, recording progress on executed postcommit methods, or using TaskFlow to execute the postcommit processing reliably and asynchronously. A similar requirement for ML2 under investigation and will likely result in a Kilo blueprint.