[autoscaler] Experimental support for local / on-prem clusters (#2678)

This adds some experimental (undocumented) support for launching Ray on existing nodes. You have to provide the head ip, and the list of worker ips.

There are also a couple additional utils added for rsyncing files and port-forward.

python/ray/autoscaler/autoscaler.py

@@ -65,6 +65,8 @@ CLUSTER_CONFIG_SCHEMA = {
             "module": (str,
                        OPTIONAL),  # module, if using external node provider
             "project_id": (None, OPTIONAL),  # gcp project id, if using gcp
+            "head_ip": (str, OPTIONAL),  # local cluster head node
+            "worker_ips": (list, OPTIONAL),  # local cluster worker nodes
         },
         REQUIRED),