ERROR: can not read leaf page from disk at serial s3 scan

[pashkinelfe]

at ed7d195
(fixed dsm leak is not relevant as in this ticket I enforce serial scan, so equivalent to current pre_merge)

Reproduction like
#345 but with no parallel scan.

set max_parallel_workers_per_gather to 0;

select count(*) from data_1;
 count
-------
  6516
(1 row)

explain select count(*) from data;
QUERY PLAN
---------------------------------------------------------------------------
 Aggregate  (cost=1355319.96..1355319.97 rows=1 width=8)
   ->  Append  (cost=0.00..1246622.83 rows=43478855 width=0)
         ->  Seq Scan on data_1  (cost=0.00..102.15 rows=4315 width=0)
         ->  Seq Scan on data_2  (cost=0.00..105.62 rows=4462 width=0)
         ->  Seq Scan on data_3  (cost=0.00..102.15 rows=4315 width=0)
         ->  Seq Scan on data_4  (cost=0.00..102.15 rows=4315 width=0)
         ->  Seq Scan on data_5  (cost=0.00..102.15 rows=4315 width=0)
         ->  Seq Scan on data_6  (cost=0.00..105.62 rows=4462 width=0)
         ->  Seq Scan on data_7  (cost=0.00..102.15 rows=4315 width=0)
         ->  Seq Scan on data_8  (cost=0.00..103.89 rows=4389 width=0)
         ->  Seq Scan on data_9  (cost=0.00..102.15 rows=4315 width=0)
.. all other 10000 partitions.

select count(*) from data;
ERROR:  can not read leaf page from disk