Fix flaky test citus_split_shard_columnar_partitioned

[hanefi]

The order of the last two rows on a table can change, and this creates a flaky test. I suggest you add an order on nodeport to make the results deterministic in these cases.

Originally posted by @hanefi in https://github.com/citusdata/citus/pull/6067#discussion_r936922972

[hanefi]

Here is the failure diff on my local:

diff --git a/src/test/regress/expected/citus_split_shard_columnar_partitioned.out b/src/test/regress/expected/citus_split_shard_columnar_partitioned.out
index b1cebc392..5a599231e 100644
--- a/src/test/regress/expected/citus_split_shard_columnar_partitioned.out
+++ b/src/test/regress/expected/citus_split_shard_columnar_partitioned.out
@@ -556,8 +556,8 @@ SET citus.shard_replication_factor TO 1;
  8999120 | colocated_partitioned_table_2020_01_01 | -2147483648   | -2127770000   | localhost |    57637
  8999121 | colocated_partitioned_table_2020_01_01 | -2127769999   | -2120000000   | localhost |    57638
  8999021 | colocated_partitioned_table_2020_01_01 | -2119999999   | 2147483647    | localhost |    57638
- 8970011 | reference_table                        |               |               | localhost |    57637
  8970011 | reference_table                        |               |               | localhost |    57638
+ 8970011 | reference_table                        |               |               | localhost |    57637
 (35 rows)

     SELECT count(*) FROM reference_table;

[onurctirtir]

Probably not applicable anymore as we don't see any more occurrences of them.