[query] one more performance improvement for getOccByNgramsOnlyFast

This time, we reduce one CTE table, this makes the query shorter,
doesn't reduce readability, maybe it will make it easier on the
postgres planner.

src/Gargantext/Database/Action/Metrics/NgramsByContext.hs

@@ -132,7 +132,8 @@ getOccByNgramsOnlyFast_withSample cId int nt ngs =
   HM.fromListWith (+) <$> selectNgramsOccurrencesOnlyByContextUser_withSample cId int nt ngs
 
 
--- Returns occurrences of ngrams in given corpus/list (for each ngram, a list of contexts is returned)
+-- | Returns occurrences of ngrams in given corpus/list (for each
+-- ngram, a list of contexts is returned)
 getOccByNgramsOnlyFast :: CorpusId
                        -> ListId
                        -> NgramsType
@@ -166,16 +167,14 @@ getOccByNgramsOnlyFast cId lId nt = do
                   FROM context_node_ngrams
                   WHERE context_id IN (SELECT context_id FROM nc)
                 ),
-                node_context_ids AS
-                ( SELECT context_id, ngrams_id, terms
+                ncids_agg AS
+                ( SELECT array_agg(DISTINCT context_id) AS agg,
+                         ngrams_id,
+                         terms
                   FROM cnnv
                   JOIN ngrams
                     ON cnnv.ngrams_id = ngrams.id
-                ),
-                ncids_agg AS
-                ( SELECT ngrams_id, terms, array_agg(DISTINCT context_id) AS agg
-                    FROM node_context_ids
-                    GROUP BY (ngrams_id, terms)
+                  GROUP BY (ngrams_id, terms)
                 ),
                 ns AS
                   (SELECT ngrams_id, terms