This is needed because in 3.6 we cannot reconcile non-cluster roles in the
bootstrap reconciliation code.
In 3.7 this is taken care of in code.

Signed-off-by: Simo Sorce <simo@redhat.com>

The playbooks were crossing byo/common boundaries for task includes.
This moves all 'common' files/tasks into the 'common' folder.

All references to it were removed in
b89c835e3235f2628b37de15713c311d1b5a4bad

All references to it were removed in
739ad9de2a11bb8c67d77641a86cfd39e4594542

In a parallel step prior to real upgrade tasks, clear out all unused
Docker images on all hosts. This should be relatively safe to interrupt
as no real upgrade steps have taken place.

Once into actual upgrade, we again clear all images only this time with
force, and after stopping and removing all containers.

Both rmi commands use a new and hopefully less error prone command to do
the removal, this should avoid missed orphans as we were hitting before.

Added some logging around the current image count before and after this
step, most of them are only printed if we're crossing the 1.10 boundary
but one does not, just for additional information in your ansible log.

We now handle the two pieces of upgrade that require a node evac in the
same play. (docker, and node itself)