np.average lacks of proper error message and docs when weights is 1D and axis is a tuple

Since version 1.7.0, the axis in np.average could be a tuple of ints. However, when weights is a 1-D array and axis is a tuple, this line of code will raise a TypeError

>>> data = np.arange(6).reshape((3, 2))
>>> data
array([[0, 1],
       [2, 3],
       [4, 5]])
>>> np.average(data, weights=[3, 1], axis=(0, 1))

   1145                 raise TypeError(
   1146                     "1D weights expected when shapes of a and weights differ.")
-> 1147             if wgt.shape[0] != a.shape[axis]:
   1148                 raise ValueError(
   1149                     "Length of weights not compatible with specified axis.")

TypeError: tuple indices must be integers or slices, not tuple

I think we should check whether the axis is a int when weights is a 1-D array and raise a TypeError like this:
TypeError: Axis must be a single int, not a tuple when shapes of a and weights differ.

And the doc about axis should be updated with something like this:

NOTE: The axis can be a tuple of ints only when weights has the same shape with the input array.

Can I create a PR to fix this?

Start by reproducing the provided example and inspect the linked axis-handling code in numpy/lib/function_base.py. Update the behavior and the axis documentation so the tuple case with 1-D weights is clearly handled, then rerun the example to confirm the resulting error and documentation match the issue.