The work done in YANGTOOLS-435 ends up lowering safety guarantees about what streamChild() actually does.

This means that if we take a class generated for a top-level Notification or and RpcInput, defeating compile-time safeguards to create an InstanceIdentifier, the codec will happily generate a YangInstanceIdentifier containing the QName of that construct.

As such, the changes we've made in YANGTOOLS-435 need to be undone and we need separate methods to enter into notification, rpc and action contexts, so that attempts at such misuse result in codec failure.