pgsql: Stabilize 026_overwrite_contrecord test

Stabilize 026_overwrite_contrecord test

On slow machines, this test can take long enough to generate WAL that a
time-based checkpoint occurs before the primary is stopped. If the
checkpoint record is written to the tail WAL segment that the test later
removes, a standby initialized from the resulting backup tries to read
the missing checkpoint record and fails with a PANIC during startup.
This caused the test to fail on buildfarm member skink.

Fix this by setting checkpoint_timeout high enough to prevent unrelated
checkpoints during the test. This follows the approach used by other
recovery tests, such as 043_no_contrecord_switch.pl, that depend on a
specific WAL layout.

Backpatch to all supported versions.

Reported-by: Alexander Lakhin <exclusion(at)gmail(dot)com>
Author: Fujii Masao <masao(dot)fujii(at)gmail(dot)com>
Discussion: https://postgr.es/m/CAHGQGwFajfjKhvNSTkTcE-wFn30p_A0_i1cvk-Q=FJhDr_5cXA@mail.gmail.com
Discussion: https://postgr.es/m/9ffdb19a-7a89-424e-925a-dd981c37f0ba@gmail.com
Backpatch-through: 14

Branch
------
REL_16_STABLE

Details
-------
https://git.postgresql.org/pg/commitdiff/66fca522c3fdb6e8c0559fd5ee44893f0732fbb8

Modified Files
--------------
src/test/recovery/t/026_overwrite_contrecord.pl | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)