Change post-filters to pre-filters in BigQueryVectorStore (#554)

BigQueryVectorStore's _create_filters are post-filters. Now that BigQuery Vector Search supports pre-filtering on stored columns (https://cloud.google.com/bigquery/docs/vector-index#stored-columns), we can change all post-filters to pre-filters.

Note that this requires storing the columns that are referenced by the pre-filter. This cannot be done in Langchain yet. If the required columns are not stored, then the pre-filters will act as post-filters. The results returned by the new post-filters will be a superset of the post-filters done currently, which can return less results.

libs/community/langchain_google_community/bq_storage_vectorstores/bigquery.py

@@ -333,9 +333,9 @@ def _create_search_query(
         full_query = f"""{embeddings_query}
         {select_clause}
         FROM VECTOR_SEARCH(
-            TABLE `{self.full_table_id}`,
+            (SELECT * FROM `{self.full_table_id}` WHERE {where_filter_expr}),
             "{self.embedding_field}",
-            (SELECT row_num, {self.embedding_field} from embeddings),
+            (SELECT row_num, {self.embedding_field} FROM embeddings),
             distance_type => "{self.distance_type}",
             top_k => {k}
         )
@@ -346,7 +346,6 @@ def _create_search_query(
         FROM (
             {full_query}
         ) AS result
-        WHERE {where_filter_expr}
         ORDER BY row_num, score
         """
         return full_query_wrapper