Fix an intermetant BF failure in 003_logical_slots.

Fix an intermetant BF failure in 003_logical_slots.

During upgrade, when pg_restore performs CREATE DATABASE, bgwriter or
checkpointer may flush buffers and hold a file handle for pg_largeobject,
so later TRUNCATE pg_largeobject command will fail if OS (such as older
Windows versions) doesn't remove an unlinked file completely till it's
open. The probability of seeing this behavior is higher in this test
because we use wal_level as logical via allows_streaming => 'logical'
which in turn set shared_buffers as 1MB and make it more probable for
bgwriter to hold the file handle.

Diagnosed-by: Alexander Lakhin
Author: Hayato Kuroda, Amit Kapila
Reviewed-by: Alexander Lakhin
Discussion: https://postgr.es/m/TYAPR01MB5866AB7FD922CE30A2565B8BF5A8A@TYAPR01MB5866.jpnprd01.prod.outlook.com