Allow openshift-ansible image to deploy to GCP

This moves all core functionality into the openshift-ansible repo, adds
the necessary equivalent entrypoint to the openshift-ansible installer
image, and ensures the dynamic inventory mechanisms in openshift-ansible
continue to work.

Notable changes from origin-gce:

* playbook extensions changed to .yml
* dynamic inventory subdirectory created to prevent accidental use
* use the custom entrypoint entrypoint-gcp for this image
* move tasks into openshift_gcp role

Also add libcloud (required for dynamic GCE lookup) and which (relied on
by gcloud).

The centos-based Dockerfile has previously been updated to pull in the python2-boto3 RPM; update the RHEL-based Dockerfile to do the same now that python2-boto3 has been made available in the appropriate repo.

There are modules in openshift-ansible that depend on boto3 (I see lib_utils/library/oo_iam_kms and lib_utils/library/oo_ec2_group). Add the dependency so that the origin-ansible container can succesfully run these modules.

The upstream forces versions on packages, and rpm -V fails when it does
not match. Also add which, used by gcloud.

The installer image will become our default cloud operations image for
tasks that overlap with ansible, and will also be the image for
provioning. We will eventually add others like awscli for support roles.

When attempting to run openshift-ansible in a kubernetes pod, we
attempted to use a configmap to mount in our inventory to run against.
However kubernetes mounts configmaps/secrets volumes through the use of
relative symlinks. The use of "cp -a" in this script then copies a
relative symlink to /tmp which is now dangling, and ansible will fail to
run.

This patch changes to just copy the file regularly resulting in a value
temp file that is not a symlink.