[sqlserver] properly close cursor, avoid leaks

We saw memory leak issues with that check. It could have been caused by
two different things:

* The fact that we are not closing cursors properly, and therefore not
  freeing properly the recordsets loaded in them
* We instantiated wrongly two different cursors in the main check loop
  letting one leak...

checks.d/sqlserver.py

@@ -232,6 +232,8 @@ def get_sql_type(self, instance, counter_name):
             except Exception, e:
                 self.log.warning("Could not get counter_name of base for metric: %s", e)
 
+        self.close_cursor(cursor)
+
         return sql_type, base_name
 
     def check(self, instance):
@@ -241,13 +243,25 @@ def check(self, instance):
         custom_tags = instance.get('tags', [])
         instance_key = self._conn_key(instance)
         metrics_to_collect = self.instances_metrics[instance_key]
-        cursor = self.get_cursor(instance)
         for metric in metrics_to_collect:
             try:
                 metric.fetch_metric(cursor, custom_tags)
             except Exception, e:
                 self.log.warning("Could not fetch metric %s: %s" % (metric.datadog_name, e))
 
+        self.close_cursor(cursor)
+
+    def close_cursor(self, cursor):
+        """
+        We close the cursor explicitly b/c we had proven memory leaks
+        We handle any exception from closing, although according to the doc:
+        "in adodbapi, it is NOT an error to re-close a closed cursor"
+        """
+        try:
+            cursor.close()
+        except Exception as e:
+            self.log.warning("Could not close adodbapi cursor\n{0}".format(e))
+
 
 class SqlServerMetric(object):
     '''General class for common methods, should never be instantiated directly