Backport of #31347.

• Explicitly remove 0 from eigenvalue signs to ensure that vh is unitary.
• Add Hermitian SVD test that has 0 as a singular value.

PR summary

Currently np.linalg.svd can return non-unitary vh due to sign 0 appearing when input array is singular.
This PR fixes the issue.

>>> import numpy as np
>>> res = np.linalg.svd(np.array([[1, 0], [0, 0]]), hermitian=True)
>>> res.Vh
array([[1., 0.],
       [0., 0.]])

AI Disclosure

I've used ChatGPT to ensure that my PR is following the guideline. No codes are generated by AI.