epmd daemon is mismanaged by puppet manifests

Detailed bug description:

 As a part of RabbitMQ installation puppet manifests start epmd daemon. E.g. master branch at the current moment:
 - https://github.com/openstack/fuel-library/blob/5726e825b69be4fa817836f94c0d5601fcaaaa50/deployment/puppet/fuel/manifests/rabbitmq.pp#L185
 - https://github.com/openstack/fuel-library/blob/5726e825b69be4fa817836f94c0d5601fcaaaa50/deployment/puppet/osnailyfacter/manifests/rabbitmq/rabbitmq.pp#L167

 But in systemd-based OSes there are actually systemd service and socket units for epmd, and this explicit call to `epmd` makes systemd dependency mechanisms unusable. As this manually run epmd now prevents systemd from listening on the same port.

Steps to reproduce:
 Try to start any systemd services that has 'Requires=epmd@0.0.0.0.socket' on master node after deployment.

Expected results:
 Service starts.

Actual result:
 Services fails to start because systemd can't satisfy it's dependency on epmd

Reproducibility:
 100%

Workaround:
 Don't use 'Requires=epmd' in systemd units

Impact:
 Switching to modern rabbitmq package can completely break deploy on systemd-enabled systems.

Additional information:
 Systemd units are slightly different on xenial and centos7. Xenial has just 'epmd.service/socket', but CentOS7 allows to specify listening address for socket like 'epmd@0.0.0.0.socket'. And plain `epmd.socket` on CentOS7 listens just on localhost.