dpt.multiply returns incorrect result when multiplying a real fp. array by a complex scalar

```
In [12]: dpt.multiply(dpt.ones(4, dtype="f4"), dpt.asarray(2j))
Out[12]: usm_ndarray([0.+2.j, 0.+2.j, 0.+2.j, 0.+2.j], dtype=complex64)

In [13]: dpt.multiply(dpt.ones(4, dtype="f4"), 2j)
Out[13]: usm_ndarray([0., 0., 0., 0.], dtype=float32)
```