Copy from ndarray (#817)

* Added blocking kernel to copy-and-cast np.ndarray -> usm_ndarray

The kernel accesses NumPy allocation using read-only buffer, hence
the blocking.

```
In [1]: import dpctl.tensor as dpt

In [2]: import numpy as np

In [3]: X = dpt.zeros((81,81,81),'d')

In [4]: %time X[:41,:41,:41] = np.random.randn(41,41,41)
CPU times: user 100 ms, sys: 8.28 ms, total: 108 ms
Wall time: 106 ms

In [5]: %time X[:41,:41,:41] = np.random.randn(41,41,41)
CPU times: user 17.8 ms, sys: 3.81 ms, total: 21.6 ms
Wall time: 20 ms
```

This used to take over 30 seconds.

* Deployed copy-and-cast kernel to copy from numpy to usm_ndarray

* Ensure we copy actual file before symlinks

Add implementation of build_package_data method for the subclass
which does not use os.chmod that does not suppot symbolic links

Author: oleksandr-pavlyk

dpctl/tensor/_copy_utils.py

@@ -78,18 +78,12 @@ def _copy_from_numpy_into(dst, np_ary):
     "Copies `np_ary` into `dst` of type :class:`dpctl.tensor.usm_ndarray"
     if not isinstance(np_ary, np.ndarray):
         raise TypeError("Expected numpy.ndarray, got {}".format(type(np_ary)))
-    src_ary = np.broadcast_to(np.asarray(np_ary, dtype=dst.dtype), dst.shape)
-    if src_ary.size and (dst.flags & 1) and src_ary.flags["C"]:
-        dpm.as_usm_memory(dst).copy_from_host(src_ary.reshape((-1,)).view("u1"))
-        return
-    if src_ary.size and (dst.flags & 2) and src_ary.flags["F"]:
-        dpm.as_usm_memory(dst).copy_from_host(src_ary.reshape((-1,)).view("u1"))
-        return
-    for i in range(dst.size):
-        mi = np.unravel_index(i, dst.shape)
-        host_buf = np.array(src_ary[mi], ndmin=1).view("u1")
-        usm_mem = dpm.as_usm_memory(dst[mi])
-        usm_mem.copy_from_host(host_buf)
+    if not isinstance(dst, dpt.usm_ndarray):
+        raise TypeError("Expected usm_ndarray, got {}".format(type(dst)))
+    src_ary = np.broadcast_to(np_ary, dst.shape)
+    ti._copy_numpy_ndarray_into_usm_ndarray(
+        src=src_ary, dst=dst, sycl_queue=dst.sycl_queue
+    )
 
 
 def from_numpy(np_ary, device=None, usm_type="device", sycl_queue=None):