v_ahb_diff: cross-product on repeated (path, segment_code) produces extra rows

[hf-mrdachner]

Problem

When (path, segment_code) appears more than once within the same (pruefidentifikator, format_version) — e.g. two "Name des Kontoinhabers" D_3192 fields under PARTIN FII — the JOIN produces a cross-product: 2 old × 2 new = 4 diff rows instead of 2.

Example: PARTIN/37000, FV2510 → FV2604

Source data: 2 rows with D_3192 in each version (identical content).

Diff result:

diff_status	old	new
unchanged	Name des Kontoinhabers	Name des Kontoinhabers
modified	Name des Kontoinhabers	Name des Kontoinhabers
modified	Name des Kontoinhabers	Name des Kontoinhabers
unchanged	Name des Kontoinhabers	Name des Kontoinhabers

Expected: 2 unchanged rows. Actual: 2 unchanged + 2 false modified.

Affected cases (cnt=2 only, non-IFTSTA)

Nachricht	Prüfis	Segment	Element
PARTIN	37000-37014	FII	D_3192 (Name des Kontoinhabers)
MSCONS	13017	(segment_group)	- (Referenzangaben)
PRICAT	27002	IMD	-

Possible fix

Use ROW_NUMBER() OVER (PARTITION BY path, segment_code ORDER BY sort_path) as an additional tiebreaker in the JOIN to prevent cross-matching repeated elements.

Related: #256, #257, #258

[hf-kklein]

Sounds like great test cases for the new id path ;)

[hf-kklein]

Addressed in #260. The cross-product issue with repeated (path, segment_code) tuples (PARTIN FII, MSCONS segment groups, PRICAT IMD) is resolved by fixing id_path construction itself rather than changing the diff JOIN key.

For these repeated elements, the #N occurrence counter fallback ensures uniqueness. Since both versions have the same repeated elements in the same order, the counter is stable across versions.

Each of these cases is tested in the new test_id_path_uniqueness_per_message_type parametrized test. Thanks for documenting these edge cases, @hf-mrdachner!