Orca falls back to planner due to cache lookup failure

[hsyuan]

Repro:

create table foo(a varchar, b oid);
explain select * from foo t1, foo t2 where t1.a::int = t2.b::int;

Output messages:

LOG:  2018-01-16 21:03:23:574312 CST,THD000,ERROR,"Lookup of object 3.26.1.0;23.1.0 in cache failed",
2018-01-16 21:03:23:574598 CST,THD000,ERROR,"Lookup of object 3.26.1.0;23.1.0 in cache failed",
LOG:  Planner produced plan :1
                                               QUERY PLAN
---------------------------------------------------------------------------------------------------------
 Gather Motion 3:1  (slice2; segments: 3)  (cost=4440.00..2797441.60 rows=2460160 width=72)
   ->  Hash Join  (cost=4440.00..2797441.60 rows=820054 width=72)
         Hash Cond: t1.a::integer = t2.b::integer
         ->  Seq Scan on foo t1  (cost=0.00..596.00 rows=16534 width=36)
         ->  Hash  (cost=2580.00..2580.00 rows=49600 width=36)
               ->  Broadcast Motion 3:3  (slice1; segments: 3)  (cost=0.00..2580.00 rows=49600 width=36)
                     ->  Seq Scan on foo t2  (cost=0.00..596.00 rows=16534 width=36)
 Settings:  optimizer=on
 Optimizer status: legacy query optimizer
(9 rows)