Currently, openshift_hosted role duplicates some logic
across separate task chains.  This commit cleans up
the openshift_hosted role and converts it to be
primarily used with include_role to give better
logic to the playbooks that utilize this role.

This commit also refactors the playbook that calls
various openshift_hosted roles into individual playbooks.
This allows more granularity for advanced users.

* [openshift_projects] Add openshift_projects role
* [openshift_hosted] hosted deployments use openshift_hosted_infra_selector if openshift_hosted_<component>_selector is not defined
* [openshift_hosted] move openshift_projects, openshift_serviceaccounts and openshift_metrics to dependencies of openshift_hosted
* [router] improve router deployment
  - add router option to force subdomain
  - add CA to router certificate options
* [registry] move registry config into openshift_hosted role
  - additional registry fixes/tweaks
  - add s3 storage support for registry
* [serviceaccount] fix up serviceaccount creation

* Move openshift_router to openshift_hosted role which will eventually
  contain registry, metrics and logging.
* Adds option for specifying an openshift_hosted_router_certificate
  cert and key pair.
* Removes dependency on node label variables and retrieves the node
  list from the API s.t. this role can be applied to any cluster with
  existing nodes. I've added an openshift_hosted playbook that occurs
  after node install to account for this.
* Infrastructure nodes are selected using
  openshift_hosted_router_selector which is based on deployment type
  by default; openshift-enterprise -> "region=infra" and online ->
  "type=infra".