IMHO, managers should not be enabled for task scheduling by default, since they cannot be properly integrated with Azure Load Balancer and made highly available at the same time (see Additional Information).
When creating a Docker Swarm, the manager node can schedule tasks.
In order to make task scheduling on managers useful for Internet-facing services, all managers would need to be part of the same backend pool as the workers. Being in the same backend pool would also mean that all nodes (workers and managers) would be part of the same Availability Set. Given unlucky circumstances, a mixed AS of workers and managers may lose the majority of managers if a fault domain goes down.
Putting managers in their own AS avoids this, but now all managers would belong to a different load balancer backend pool than the worker nodes, which exposes all service instances on the managers nodes on a different public IP address as well.
Steps to reproduce the behavior
Create a Docker Swarm using the 1.12.1-beta 5 ARM template.