Fix OrderedNamespaceSet.__setitem__: int remove-before-add and slice exhausted iterator

[zrgt]

Fixes #498
Fixes #494

Changes

__setitem__ int (index replacement)

• Remove old item from backends before adding new item — prevents false AASConstraintViolation when new item shares an attribute (e.g. id_short) with the item being replaced
• Rollback: re-adds old item if the subsequent add fails

__setitem__ slice

• Replace self._order[s] = new_items with self._order[s] = successful_new_items — new_items is an exhausted islice iterator; assigning it back to the list slice wrote an empty sequence, silently discarding all new items

Tests

• test_ordered_namespaceset_int_setitem_preserves_index: builds [p0, old, p2], replaces index 1 with new (same id_short), asserts result is [p0, new, p2] — no index shift, no AASConstraintViolation
• test_ordered_namespaceset_slice_setitem_preserves_order: replaces a 2-item slice, verifies new items appear at correct positions

[s-heppner]

Doesn't the new (and old) implementation change the item's index? Doesn't __setitem__ usually keep the index?

I think the method __set_item__ should actually make use of Referable.update_from(), which makes sure they don't accidentally change immutable attributes.

[zrgt]

On index preservation: The index is preserved — self._order[s] = o is a list assignment that places the new item exactly at position s, leaving all surrounding elements unchanged. Added test_ordered_namespaceset_int_setitem_preserves_index to make this explicit: builds [p0, old, p2], replaces index 1 with new (same id_short as old), and asserts the result is [p0, new, p2].

On update_from(): Using update_from() would change the semantics — the caller passes a new object o intending it to be the item at index s, but update_from() mutates the existing object in place and discards o. If the caller holds a reference to o expecting it to be in the collection after ns[1] = o, they would be surprised. Standard Python __setitem__ semantics (e.g. list.__setitem__) replace the item at the index with the new object, which is what we implement here.

[zrgt]

Merged #516 (slice fix: exhausted iterator) into this PR. Both int and slice fixes are now covered here. PR #516 has been closed.