Description of problem:
Edit installer.cfg.yml to have a dedicated etcd host, and then trigger an install/scaleup, the installation summary output is not correct.
*** Installation Summary ***

Hosts:
- openshift-118.x.x.x
  - OpenShift master
  - OpenShift node (Unscheduled)
  - Etcd (Embedded) //should not be etcd
  - Storage
- openshift-136.x.x.x
  - OpenShift node (Dedicated)
- openshift-116.lab.eng.nay.redhat.com  //should be etcd role

Total OpenShift masters: 1
Total OpenShift nodes: 2


Version-Release number of selected component (if applicable):
atomic-openshift-utils-3.4.13-1.git.0.ff1d588.el7.noarch


How reproducible:
always

Steps to Reproduce:
1.Edit installer.cfg.yml
#cat /root/.config/openshift/installer.cfg.yml
<--snip-->
deployment:
  ansible_ssh_user: root
  hosts:
  - connect_to: openshift-x.x.x
    hostname: 192.168.2.4
    ip: 192.168.2.4
    public_hostname: 10.66.147.118
    public_ip: 10.66.147.118
    roles:
    - master
    - node
    - storage
  - connect_to: openshift-x.x.x
    hostname: 192.168.2.53
    ip: 192.168.2.53
    node_labels: '{''region'': ''infra''}'
    public_hostname: 10.66.147.136
    public_ip: 10.66.147.136
    roles:
    - node
  - connect_to: openshift-x.x.x
    hostname: 192.168.2.46
    ip: 192.168.2.46
    public_hostname: 10.66.147.116
    public_ip: 10.66.147.116
    roles:
    - etcd

2.trigger an install
# atomic-openshift-installer -c /root/.config/openshift/installer.cfg.yml install


Actual results:
*** Installation Summary ***

Hosts:
- openshift-118.x.x.x
  - OpenShift master
  - OpenShift node (Unscheduled)
  - Etcd (Embedded) //should not be etcd
  - Storage
- openshift-136.x.x.x
  - OpenShift node (Dedicated)
- openshift-116.lab.eng.nay.redhat.com  //should be etcd role

Total OpenShift masters: 1
Total OpenShift nodes: 2

Expected results:
The printed summary should be consistent with installer.cfg.yml.