gh-158788: Preserve metaclass __call__ in vectorcall optimization

[cocolato]

The vectorcall optimization for class calls can bypass a custom metaclass's __call__ after a call site specializes on a built-in class.

In CALL_BUILTIN_CLASS, the runtime guard accepts any class with a non-NULL tp_vectorcall, and the call handler invokes that pointer directly.

Require the class's metaclass to be exactly type in the runtime guard, matching the optimizer's existing assumption. Move the existing metaclass check before the immutable-type branch in specialize_class_call() so immutable classes with custom metaclasses also use the generic call path.

• Issue: Vectorcall optimization can bypass a custom metaclass's __call__ #158788