GeezerVison

Problem

• del obj[key] on any reflected type with a settable indexer crashed the process.
• CPython uses the same mp_ass_subscript slot for assignment and deletion, with a null value for del.
• mp_ass_subscript_impl never checked for the null and forwarded it into PyTuple_SetItem, which threw across the native boundary.
• ArrayObject.mp_ass_subscript had the same hole and died with an access violation.
• Official pythonnet fixed this in __delitem__ for IList<T> and IDictionary<K,V> pythonnet/pythonnet#2533; this fork predates it.

Changes

• ClassBase: check v.IsNull first and route to the new DeleteItemImpl.
• DeleteItemImpl: TypeError when the type has no deleter or the key is a tuple; otherwise call the deleter through the binder; false from Remove becomes KeyError.
• Indexer: new DeleterBinder, resolved lazily on the first del from IDictionary<K,V>.Remove(K), else IList<T>.RemoveAt(int). Uses the interface method, so explicit implementations like ConcurrentDictionary.Remove work.
• ArrayObject: TypeError on a null value.
• Tests: nine pytest cases, a throwing Remove fixture, and an embed test fixture.

Behaviour

Verification

• pytest: 477 passed, 1 pre-existing environmental failure (test_explicit_assembly_load).
• Embed tests (Debug): 3 passed. Each new test crashes on master.
• Lean with the patched DLL: a Python algorithm doing del self.runtime_statistics[key] now completes; 324/324 Python regression algorithms and 16600/16600 Python unit tests pass.

Follow-up

• Stubs: Emit __delitem__ on IDictionary and IList quantconnect-stubs-generator#100 adds __delitem__ on IDictionary and IList. Ship it with or after this release.