Summary:
Fix is_dict_class detection to use the fast decode path. This is performance gap, not a correctness bug.

Description:

Context

is_dict_class is computed with Py_IS_TYPE(options->document_class, &PyDict_Type). document_class is a class (a type object), and Py_IS_TYPE(x, T) is Py_TYPE == T. So we are comparing{{Py_TYPE(&PyDict_Type) == &PyDict_Type}}, which is false (it is PyType_Type). So is_dict_class was always false, forcing every document decode through the generic PyObject_SetItem path instead of the fast PyDict_SetItem path, adding object churn.

Definition of done

• Compute is_dict_class with an identity comparison: options->document_class == (PyObject*)&PyDict_Type.
• test_bson.py and test_bson_corpus.py pass.

Pitfalls

• document_class is a class, not an instance, so Py_IS_TYPE is the wrong check; use identity equality.