fix(eam): terminate instances that failed before start without SRM

SRM sends failed_before_start when it rejects a deploy command before creating anything. On that path it writes no service order or instance rows. OEG marked the instances FAILED but still sent service.terminate on DELETE. SRM then hit a ForeignKeyViolationError on service_order, sent no completion, and the instance stayed terminating until the 30-minute reconcile deadline (SRM-3).

handle_operation_status now flags the deploy operation with failed_before_start. On DELETE /appinstances/{id} and DELETE /deployments/{id}, FAILED instances of a flagged deploy are set to TERMINATED locally and the subscription callback is sent. No terminate command is published for them. A deployment with nothing left for SRM settles through the existing no-live-instances path.

The flag is per operation. That is safe for multi-zone deployments because SRM rejects the whole command before it creates the service order. A plain "failed" status still goes through SRM terminate.