Fixed a bug in dpt.arange

A bug would manifest itself if the starting value is outside of
range of the array data type.

Fixed it by coercing the starting value to the data type, like
is currently done for the increment.

The test was modified to cover such an input.

Author: oleksandr-pavlyk

dpctl/tensor/_ctors.py

@@ -547,7 +547,8 @@ def arange(
     )
     _step = (start + step) - start
     _step = dt.type(_step)
-    hev, _ = ti._linspace_step(start, _step, res, sycl_queue)
+    _start = dt.type(start)
+    hev, _ = ti._linspace_step(_start, _step, res, sycl_queue)
     hev.wait()
     return res
 

dpctl/tests/test_usm_ndarray_ctor.py

@@ -985,11 +985,13 @@ def test_arange(dt):
     elif np.issubdtype(dt, np.complexfloating):
         assert complex(X[47]) == 47.0 + 0.0j
 
-    X1 = dpt.arange(4, dtype=dt, sycl_queue=q)
-    assert X1.shape == (4,)
+    # choose size larger than maximal value that u1/u2 can accomodate
+    sz = int(np.iinfo(np.int16).max) + 1
+    X1 = dpt.arange(sz, dtype=dt, sycl_queue=q)
+    assert X1.shape == (sz,)
 
-    X2 = dpt.arange(4, 0, -1, dtype=dt, sycl_queue=q)
-    assert X2.shape == (4,)
+    X2 = dpt.arange(sz, 0, -1, dtype=dt, sycl_queue=q)
+    assert X2.shape == (sz,)
 
 
 @pytest.mark.parametrize(