Support for ndarray subclassing via `view(type=...)`

[abagusetty]

dpnp currently does not support creating subclasses of dpnp.ndarray using the .view(type=CustomClass) pattern that is standard in NumPy and CuPy. This limitation makes it difficult to port NumPy/CuPy code that relies on ndarray subclassing to dpnp.

Reproducer:

import dpnp

class MySubclass(dpnp.ndarray):
    pass

arr = dpnp.array([1, 2, 3])
sub = arr.view(type=MySubclass)  # NotImplementedError!

output:

Traceback (most recent call last):
  File "/home/abagusetty/gpu4pyscf-testing/gpu4pyscf/test_dpnp_subclass_v1.py", line 7, in <module>
    sub = arr.view(type=MySubclass)  # NotImplementedError!
  File "/home/abagusetty/gpu4pyscf-testing/dpnp/dpnp/dpnp_array.py", line 2355, in view
    raise NotImplementedError(
NotImplementedError: Keyword argument `type` is supported only with default value ``None``, but got <class '__main__.MySubclass'>.

Works fine with numpy & cupy:

import numpy as np
#import cupy as np

class MySubclass(np.ndarray):
    pass

arr = np.array([1, 2, 3])
sub = arr.view(MySubclass)
sub.custom = "test"

print(type(sub))           # <class '__main__.MySubclass'>
print(sub.custom)          # test
print(hasattr(sub, 'custom'))  # True

output:

<class '__main__.MySubclass'>
test
True

Use case:
This feature is essential for scientific computing libraries that need to:

• Attach metadata to arrays: Track provenance, units, coordinate systems, etc.
• Maintain compatibility with NumPy/CuPy code: Many libraries use subclassing for array tagging (eg. gpu4pyscf)