Indexing maskedarray with ... gives masked #8684

>>> X = np.array([(1,[2,3])], dtype=[('',int),('',float,(2,))])
>>> Y = np.ma.array(X, mask=[(True, False)])
>>> X[0]
(1, [ 2.,  3.])
>>> Y[0]
(--, [2.0, 3.0])
>>> X[0,...]
array((1, [ 2.,  3.]),
      dtype=[('f0', '<i4'), ('f1', '<f8', (2,))])
>>> Y[0,...]
masked