Optmize read acl update by not rebuilding the full read_acls table

The nx_update_read_acls rebuild the read_acls table that contains all possible read acls, then it update the read acls for new documents:

RAISE INFO 'nx_rebuild_read_acls truncate hierarchy_read_acl';
TRUNCATE TABLE hierarchy_read_acl;
RAISE INFO 'nx_rebuild_read_acls update acl map';
INSERT INTO hierarchy_read_acl
SELECT id, md5(nx_get_read_acl(id))
FROM (SELECT id FROM hierarchy WHERE isproperty='f') AS uids;

This is fine if you have few dozen of acls, with thousands of acls it takes time.
A simple optimization is to rebuild this table only if we have new entry with an acls.