I've been made aware that ETRS89 to MGI was semi broken since the EPSG
update that added a ETRS89-AUT [2002] datum and migrated the 'MGI to
ETRS89 (8)' transformation using the last released grid AT_GIS_GRID_2021_09_28
to be 'MGI to ETRS89-AUT [2002]' and thus leaving user with the older
transformation 'MGI to ETRS89 (5)'.
This is just one manifestation of a problem that can be seen with ETRS89
in the Netherlands, Romania, etc.
Until EPSG perhaps figures out a plan to restore cleanly backward
compatibility (they have been made aware of those issues), I've devised
the following s/hack/strategy that seems to work (TM):

When looking for ETRS89 (EPSG:4258 the datum ensemble) to XXXX
transformations, also query the ``alias_name`` table for old names
of transformations that start with 'ETRS89 to XXXX' or
'XXXX to ETRS89', and are aliases to 'ETRS89-YYY to XXXX[something]'
or 'XXXX to ETRS89-YYY[something]' newly named transformation. And then
consider those 'migrated' transformations to be also used for plain
ETRS89, because they used to be used for it!

No new test cases, but actually reverting to old reference values before
commit 047e5db (update to EPSG 12.033)
that changed them.