Fixed length computations, added test

```
In [4]: import numpy

In [5]: numpy.arange(-5, stop=10**5, step=2.7, dtype=numpy.int64).shape
Out[5]: (37039,)

In [6]: import dpctl.tensor as dpt

In [7]: dpt.arange(-5, stop=10**5, step=2.7, dtype=numpy.int64).shape
Out[7]: (50003,)
```

Now lengths are consistent.

Author: oleksandr-pavlyk

dpctl/tensor/_ctors.py

@@ -491,13 +491,14 @@ def _round_for_arange(tmp):
 def _get_arange_length(start, stop, step):
     "Compute length of arange sequence"
     span = stop - start
-    if type(step) in [int, float] and type(span) in [int, float]:
+    if hasattr(step, "__float__") and hasattr(span, "__float__"):
         return _round_for_arange(span / step)
     tmp = span / step
-    if type(tmp) is complex and tmp.imag == 0:
+    if hasattr(tmp, "__complex__"):
+        tmp = complex(tmp)
         tmp = tmp.real
     else:
-        return tmp
+        tmp = float(tmp)
     return _round_for_arange(tmp)
 
 
@@ -553,7 +554,7 @@ def arange(
         allow_bool=False,
     )
     try:
-        tmp = _get_arange_length(start_, stop_, step_)
+        tmp = _get_arange_length(start, stop, step)
         sh = int(tmp)
         if sh < 0:
             sh = 0

dpctl/tests/test_usm_ndarray_ctor.py

@@ -1052,13 +1052,18 @@ def test_arange_mixed_types():
     assert x.shape[0] == 3
     assert int(x[1]) == 99 + int(x[0])
 
-    x = dpt.arange(+2.5, stop=200, step=100, dtype="int32", sycl_queue=q)
+    x = dpt.arange(+2.5, stop=200, step=100, dtype="int32", device=x.device)
     assert x.shape[0] == 2
     assert int(x[1]) == 100 + int(x[0])
 
-    x = dpt.arange(0, stop=np.float32(504), step=100, dtype="f4")
+    _stop = np.float32(504)
+    x = dpt.arange(0, stop=_stop, step=100, dtype="f4", device=x.device)
     assert x.shape == (6,)
 
+    # ensure length is determined using uncast parameters
+    x = dpt.arange(-5, stop=10**2, step=2.7, dtype="int64", device=x.device)
+    assert x.shape == (39,)
+
 
 @pytest.mark.parametrize(
     "dt",