[elastic] improve hostname resolution

When the `cluster_stats` parameter is enabled in the user configuration,
the Datadog Agent's Elasticsearch check attempts to resolve and
attach with the corresponding metrics, the hostname of the different
nodes.

On Amazon Elasticsearch v1.5.2, the node objects
returned by the `_cluster/nodes/stats?all=true` endpoint returns the
following keys:
```
root@xxxx:~# at node_stats.json| jq '.nodes[] | select(.name ==
"${hostname}") | keys
[
  "breakers",
  "fs",
  "indices",
  "jvm",
  "name",
  "os",
  "process",
  "thread_pool",
  "timestamp"
]
```

`name` contains the hostname.

checks.d/elastic.py

@@ -507,12 +507,20 @@ def _process_stats_data(self, nodes_url, data, stats_metrics, config):
         cluster_stats = config.cluster_stats
         for node_name in data['nodes']:
             node_data = data['nodes'][node_name]
-            # On newer version of ES it's "host" not "hostname"
-            node_hostname = node_data.get(
-                'hostname', node_data.get('host', None))
 
-            # Override the metric hostname if we're hitting an external cluster
-            metric_hostname = node_hostname if cluster_stats else None
+            metric_hostname = None
+
+            # Resolve the node's hostname
+            if cluster_stats:
+                for k in ['hostname', 'host', 'name']:
+                    if k in node_data:
+                        metric_hostname = node_data[k]
+                        break
+                else:
+                    self.log.warning(
+                        u"Unable to resolve the hostname associated to the node name '%s'",
+                        node_name
+                    )
 
             for metric, desc in stats_metrics.iteritems():
                 self._process_metric(