Incremental Sort crash on a rescan that becomes bounded

Hi hackers,

This fails an assertion in tuplesort_set_bound():

set enable_seqscan = off;
set enable_bitmapscan = off;
select o.n, count(*) from (values (null::int), (3)) o(n),
lateral (select unique1 from tenk1 where hundred < 2
order by hundred, unique1 limit o.n) s
group by o.n;

Incremental Sort creates its sort states on the first scan and reuses
them on rescan. They only allow a bound if the first scan was bounded,
so a later scan that sets a bound fails the assertion. Since 17 such a
state uses a Bump context. A bounded sort frees the tuples it throws
away, which Bump can't do. So release builds can crash too.

The attached patch ends the sort state on such a rescan. The second
patch is the version for 15.

Thanks,
Shihao