Add partOf depth-2 recursion test cases (#155)

[BIMvoice]

Adds conformance coverage for a gap that let a real implementation bug survive.

The gap

Of the 33 official partOf conformance cases, only 3 exercise same-relation recursion beyond depth 1, and none covers IFCRELASSIGNSTOGROUP or IFCRELCONTAINEDINSPATIALSTRUCTURE beyond a direct relation. One of those 3, pass-the_containment_can_be_indirect_1_2, is in fact a pure IFCRELAGGREGATES chain despite its name, so containment recursion was not exercised anywhere in the suite.

Why it matters

That is exactly where IfcOpenShell's IfcTester had a bug. Its group and containment branches inspected only the element's single direct relation, so an element two hops away silently passed a PROHIBITED requirement, reporting compliance while the relationship existed. Fixed in IfcOpenShell#9272.

Documentation/UserManual/partof-facet.md on development states the rule these cases pin:

When the relation parameter is not specified, then all 6 are to be considered (recursively) to identify containing entities, otherwise only the specified relation type should be considered (also recursively).

Note that file differs between development and ver/1.1.x; the sentence above is from development.

What is added

Two pass and fail pairs, each pair sharing a two-hop chain with distinct intermediate and target classes at every hop so an entity type collision cannot mask the result:

• IFCRELASSIGNSTOGROUP: pass-a_group_relationship_may_be_indirect and fail-a_prohibited_group_relationship_detects_an_indirect_group
• IFCRELCONTAINEDINSPATIALSTRUCTURE: pass-a_containment_relationship_may_be_indirect and fail-a_prohibited_containment_relationship_detects_an_indirect_container

Verification

Run against IfcTester both before and after the fix above:

pre-fix   the 4 new cases mismatch their own filename; the 33 existing cases are unaffected
post-fix  0 of 37 mismatch

So the cases demonstrably catch the defect, and demonstrably do not disturb the existing suite.

All 4 .ids files validate against Schema/ids.xsd, checked with xmlschema. A control file using relation="NOTAVALIDRELATION" is correctly rejected, confirming the validator discriminates rather than passing anything.

Correction: I originally wrote that CreateTestCases could not be run. That was wrong. The build needs the .NET 8 runtime specifically, because Nuke uses BinaryFormatter, which .NET 9 removed. With that installed it runs in about a second. These .ids files are still hand-authored; I am happy to regenerate them through the official build on request.

Scope

This does not address the separate "require a direct relationship" question also raised on #155, which is still open. #155 is assigned to @CBenghi, so this is offered as a starting point to take, change or discard.

This PR was created with the assistance of an AI coding tool.