Ensure PG 16+ generate same result as PG15 and before

[nyurik]

It seems there might be some cases when Martin does not generate the same result in PG 16 as in versions before. More importantly, sometimes martin-cp generates a different number of tiles from it too. We must get a very clear understanding of why it differs, and try to make them the same again.

see also #1642 for a great analysis of this issue by @sharkAndshark

[sharkAndshark]

Different result with same table and query

Looks like this is an upstream issue. I've post an issue there.

To verify and duplicate this:

1. Update the PostGIS version (postgis/postgis:14-3.3 or postgis/postgis:14-3.5) in docker-compose.yml file
2. just start
3. Run the query with zxy == 1 1 1 or 1 1 0 and watch the results.

SELECT
      ST_AsMVT(tile, 'table_source', 4096, 'geom')
    FROM (
      SELECT
        ST_AsMVTGeom(
            ST_Transform(ST_CurveToLine("geom"::geometry), 3857),
            ST_TileEnvelope($1::integer, $2::integer, $3::integer),
            4096, 64, true
        ) AS geom
        , "gid"
      FROM
        "public"."table_source"
      WHERE
        "geom" && ST_Transform(ST_TileEnvelope($1::integer, $2::integer, $3::integer, margin => 0.015625), 4326)
      
    ) AS tile;

I jsut paste the results here in case someone need it:

z	x	y	pg_version	source	st_asmvt	is_empty
1	1	1	postgis/postgis:14-3.3	table_source	(BLOB) 2 bytes	yes
1	1	0	postgis/postgis:14-3.3	table_source	(BLOB) 2 bytes	yes
1	1	1	postgis/postgis:14-3.5	table_source	(BLOB) 607 bytes	no
1	1	0	postgis/postgis:14-3.5	table_source	(BLOB) 1.18 KB	no

[sharkAndshark]

Different features order with various PostGIS versions

I think we could pay less attention on this as there is no guarantee the orders should be same without an order by.
Just update the test.sh to ignore the order with ogrmerge of gdal. I've tried and it works well.