usm_ndarray must support array asignment

[oleksandr-pavlyk]

usm_ndarray must defined __setitem__.

[oleksandr-pavlyk]

This has been closed with merge of #586 :

In [1]: import dpctl.tensor as dpt, numpy as np

In [2]: X = dpt.usm_ndarray((3, 4), dtype="d")

In [3]: for i0 in range(3):
   ...:     for i1 in range(4):
   ...:         X[i0, i1] = 2.0/(i1 + i0 + 1.0)  # assign Python scalars
   ...:

In [4]: dpt.to_numpy(X)
Out[4]:
array([[2.        , 1.        , 0.66666667, 0.5       ],
       [1.        , 0.66666667, 0.5       , 0.4       ],
       [0.66666667, 0.5       , 0.4       , 0.33333333]])

In [5]: Y = dpt.usm_ndarray((3, 4), dtype="f")

In [6]: Y[:] = X  # from usm_ndarray to usm_ndarray with type coercion

In [7]: dpt.to_numpy(Y)
Out[7]:
array([[2.        , 1.        , 0.6666667 , 0.5       ],
       [1.        , 0.6666667 , 0.5       , 0.4       ],
       [0.6666667 , 0.5       , 0.4       , 0.33333334]], dtype=float32)

In [8]: Y[:] = np.full((3,4), 2)  # from numpy array

In [9]: dpt.to_numpy(Y)
Out[9]:
array([[2., 2., 2., 2.],
       [2., 2., 2., 2.],
       [2., 2., 2., 2.]], dtype=float32)