* node attributes 'py:module', 'py:class' is provided from environment (e.x. currentmodule directive is provided on previous line, not on node)
* Because translation nodes are parsed in independently, 'provided attributes by environment' are not exist.
* For a pending_xref nodes, all attributes of a translated xref node should be same as original xref node in current test cases.
* For the above reasons, I overwrite all attributes from origonal node to new translated xref node.

The NumpyDoc extension that is developed by the Numpy folks does a lot
of extra work interpreting See Also sections. It assumes that the contents
of the See Also will always be references to other functions/classes and
it tries to deduce what is being referenced.

I've ported their implementation for See Also sections written in the
Numpy style. There is NO extra interpretation done for See Also sections
that are written using the Google style.