[Gem] Fixes ScrollHelper

Fixes #2556

The iteration implementation has been fixed. There was a bug where an
additional search (with scroll) request was made to Elasticsearch for
each resulting hit. It was rewritten so that the docs are retrieved as
needed and the Helper instance doesn't store documents internally,
with big savings in memory and requests to Elasticsearch.

Author: picandocodigo

elasticsearch/lib/elasticsearch/helpers/scroll_helper.rb

@@ -43,12 +43,8 @@ def initialize(client, index, body, scroll = '1m')
       # @yieldparam document [Hash] yields a document found in the search hits.
       #
       def each(&block)
-        @docs = []
-        @scroll_id = nil
-        refresh_docs
-        for doc in @docs do
-          refresh_docs
-          yield doc
+        until (docs = results).empty?
+          docs.each(&block)
         end
         clear
       end
@@ -70,25 +66,19 @@ def results
       #
       def clear
         @client.clear_scroll(body: { scroll_id: @scroll_id }) if @scroll_id
-        @docs = []
+        @scroll_id = nil
       end
 
       private
 
-      def refresh_docs
-        @docs ||= []
-        @docs << results
-        @docs.flatten!
-      end
-
       def initial_search
         response = @client.search(index: @index, scroll: @scroll, body: @body)
         @scroll_id = response['_scroll_id']
         response['hits']['hits']
       end
 
       def scroll_request
-        @client.scroll(body: {scroll: @scroll, scroll_id: @scroll_id})['hits']['hits']
+        @client.scroll(body: { scroll: @scroll, scroll_id: @scroll_id })['hits']['hits']
       end
     end
   end