The problem in CoreTest.testMahalanobis is that the generated covariance matrix is not invertible, because the random matrix has too few samples.

Consider 3 vectors in 3D space. They all lie on a plane, so they can be expressed as their mean + 2 linearly independent vectors. This makes their covariance matrix singular. The same applies to n ≠ 3. You can verify this by observing that in

import numpy as np
n = 10
a = np.random.uniform(-128, 128, (n, n))
print(np.linalg.eigvals(np.cov(a)))

you will for any n always get a zero eigenvalue (actually close to zero due to numerical errors).

The inverses in #24348 are different across platforms because, well, the inverse doesn't exist up to numerical errors. The fix is to add one more random sample to make the covariance matrix non-singular.

This PR resolves #24348

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake