Interpreter uses `DK_IS_UNICODE` where it should use `dk_kind == DICT_KEYS_UNICODE`

[colesbury]

We are unintentionally specializing LOAD_GLOBAL for globals and module dictionaries that have "split" keys (i.e., keys from a class instance's __dict__).

For example:

cpython/Python/specialize.c

Lines 1283 to 1286 in 6e625f8

	if (!DK_IS_UNICODE(globals_keys)) {
	SPECIALIZATION_FAIL(LOAD_GLOBAL, SPEC_FAIL_LOAD_GLOBAL_NON_STRING_OR_SPLIT);
	goto fail;
	}

This doesn't actually affect the behavior, because split keys have NULL me_value entries and we de-opt in that case:

cpython/Python/bytecodes.c

Lines 1816 to 1817 in 6e625f8

	PyObject *res_o = FT_ATOMIC_LOAD_PTR_RELAXED(entries[index].me_value);
	DEOPT_IF(res_o == NULL);

Linked PRs

• gh-142908: Don't use DK_IS_UNICODE in interpreter #142909