Fix NamespaceSet.pop() removes item from all backends

[zrgt]

Summary

NamespaceSet.pop() used popitem() on the first backend only, leaving stale entries in all remaining backends. Any subsequent add() of an item with the same semantic_id raised a false AASConstraintViolation.

Fix iterates all backends and removes the popped item's key from each.

Fixes #496

Test plan

• test_namespaceset_pop_removes_from_all_backends — verifies pop then re-add with same semantic_id succeeds
• Full test.model.test_base suite passes (53 tests)

[s-heppner]

Wait, this is nonsensical though, isn't it? Namespaces only ever identify via one attribute? I think here the class may have been misunderstood, which might mean, we did not implement them cleanly enough, but this is not a bug, as far as I can see.

[zrgt]

Thanks for the review, @s-heppner!

You're right that current production code uses only one attribute per NamespaceSet. However, NamespaceSet._backend is typed as Dict[str, Tuple[Dict, bool]] — the class is explicitly designed to support multiple uniqueness attributes. The existing test helpers ExampleNamespaceReferable.set1 and ExampleOrderedNamespace.set1 (with both id_short + semantic_id) were deliberately created to exercise this general case, showing multi-attribute support was always intentional.

A pop() that silently leaves stale entries in registered backends is a real correctness bug for the class — the fix is a no-op for single-attribute callers.