Upgrade nova-api/scheduler/conductor packages at step3 not step2.

Upgrade nova-api/scheduler/conductor packages at step3 not step2.

The nova-api, nova-scheduler nova-conductor packages are updated during
step2.  The package upgrade trigger a restart of the service which fails
and is constantly retried by systemd:

    Feb 24 12:34:24 centos-7-2-node-rax-iad-7463943-440549 systemd[1]: Failed to start OpenStack Nova Scheduler Server.
    Feb 24 12:34:24 centos-7-2-node-rax-iad-7463943-440549 systemd[1]: Unit openstack-nova-scheduler.service entered failed state.
    Feb 24 12:34:24 centos-7-2-node-rax-iad-7463943-440549 systemd[1]: openstack-nova-scheduler.service failed.
    Feb 24 12:34:24 centos-7-2-node-rax-iad-7463943-440549 systemd[1]: openstack-nova-scheduler.service holdoff time over, scheduling restart.

We eventually reach timeout.  We use
https://github.com/openstack/tripleo-heat-templates/blob/master/puppet/services/tripleo-packages.yaml#L44-L46
to upgrade existing packages.

Add a note to the README.rst to make people aware of the general upgrade
done at step3 and limit its usage to new package for individual service.

Change-Id: I13b51bcfe0c98034944613f7e1c3f0168cd4de76
Closes-Bug: #1667728
(cherry picked from commit 5593877817b332bd7a650f051cff5f5988ab1e54)