Return unitary vh from dpnp.linalg.svd for singular Hermitian inputs

[antonwolfy]

dpnp.linalg.svd(..., hermitian=True) could return a non-unitary vh when the input array is singular. In the Hermitian path, the sign of each eigenvalue is factored out of the eigenvectors into vh. The sign was computed with dpnp.sign(s), which returns 0 for zero eigenvalues, zeroing out the corresponding rows of vh and breaking unitarity.

The fix replaces dpnp.sign(s) with dpnp.copysign(1.0, s), which always yields ±1, mirroring the change included into NumPy 2.5.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• Have you checked performance impact of proposed changes?
• Have you added documentation for your changes, if necessary?
• Have you added your changes to the changelog?

[github-actions]

View rendered docs @ https://intelpython.github.io/dpnp/pull/2954/index.html

[coveralls]

coverage: 78.092%. remained the same — return-unitary-vh-from-svd into master

[github-actions]

Array API standard conformance tests for dpnp=0.21.0dev1=py313h509198e_12 ran successfully.
Passed: 1357
Failed: 3
Skipped: 16

[vlad-perevezentsev]

LGTM
Thank you @antonwolfy