Fix regression: not respecting interval in read_with_poll (#343)

* Fix regression: not respecting interval in read_with_poll

The regression happened in the port to PL/pgSQL.

Division of integers on postgres yields integer results. Our unit conversion from
milliseconds to seconds didn't consider that, then returning wrong results.
This caused `pgmq.read_with_poll` to sleep 0s instead of `poll_interval_ms`
if `poll_interval_ms` was under 1000, using 100% of CPU.

The fix is simply converting `poll_interval_ms` to `numeric` before doing the
division, which produces a `numeric` result as originally intended.

* Fix tests

pgmq-extension/sql/pgmq--1.4.4--1.5.0.sql

@@ -97,7 +97,7 @@ BEGIN
       IF FOUND THEN
         RETURN;
       ELSE
-        PERFORM pg_sleep(poll_interval_ms / 1000);
+        PERFORM pg_sleep(poll_interval_ms::numeric / 1000);
       END IF;
     END LOOP;
 END;

pgmq-extension/sql/pgmq.sql

@@ -149,7 +149,7 @@ BEGIN
       IF FOUND THEN
         RETURN;
       ELSE
-        PERFORM pg_sleep(poll_interval_ms / 1000);
+        PERFORM pg_sleep(poll_interval_ms::numeric / 1000);
       END IF;
     END LOOP;
 END;

pgmq-extension/test/expected/base.out

@@ -434,7 +434,7 @@ SELECT ARRAY(
 
 -- Read with poll will poll until the first message is available
 SELECT clock_timestamp() AS start \gset
-SELECT msg_id = :msg_id1 FROM pgmq.read_with_poll('test_read_queue', 10, 5, 5, 100);
+SELECT msg_id = :msg_id1 FROM pgmq.read_with_poll('test_read_queue', 10, 5, 6, 100);
  ?column? 
 ----------
  t

pgmq-extension/test/sql/base.sql

@@ -172,7 +172,7 @@ SELECT ARRAY(
 
 -- Read with poll will poll until the first message is available
 SELECT clock_timestamp() AS start \gset
-SELECT msg_id = :msg_id1 FROM pgmq.read_with_poll('test_read_queue', 10, 5, 5, 100);
+SELECT msg_id = :msg_id1 FROM pgmq.read_with_poll('test_read_queue', 10, 5, 6, 100);
 SELECT clock_timestamp() - :'start' > '3 second'::interval;
 
 -- test_purge_queue