Improve PostgreSQL replication lag detection (#395)

In some cases master can show pg_last_xact_replay_timestamp() from past,
which can cause the exporter to show ever-growing value for the lag.

By checking if the instance is in recovery we can avoid reporting some
huge number for master instance.

Author: jsosic

queries.yaml

@@ -1,5 +1,5 @@
 pg_replication:
-  query: "SELECT EXTRACT(EPOCH FROM (now() - pg_last_xact_replay_timestamp())) as lag"
+  query: "SELECT CASE WHEN NOT pg_is_in_recovery() THEN 0 ELSE GREATEST (0, EXTRACT(EPOCH FROM (now() - pg_last_xact_replay_timestamp()))) END AS lag"
   master: true
   metrics:
     - lag: