faster by created for observer communitites

Original implementation was very slow (~70s) when the observer wasn't subscribed to communitities, because it had to search entire hive_posts table in order to determine there were no posts to return. This algorithm picks most recent _limit posts from each community that observer is subscribed to, then sorts then by most recent.

hive/db/sql_scripts/upgrade/upgrade_table_schema.sql

@@ -73,10 +73,12 @@ drop index if exists hive_subscriptions_community_idx;
 drop index if exists hive_votes_post_id_idx;
 drop index if exists hive_votes_voter_id_idx;
 drop index if exists hive_votes_last_update_idx;
+drop index if exists hive_posts_community_id_idx;
 
 CREATE INDEX IF NOT EXISTS hive_posts_cashout_time_id_idx ON hive_posts (cashout_time, id);
 CREATE INDEX IF NOT EXISTS hive_posts_updated_at_idx ON hive_posts (updated_at DESC);
 CREATE INDEX IF NOT EXISTS hive_votes_block_num_idx ON hive_votes (block_num);
+CREATE INDEX IF NOT EXISTS hive_posts_community_id_id_idx ON hive_posts (community_id, id DESC);
 
 DO
 $BODY$