Fix GPU mean validation to honor mean_tol

[Copilot]

GPU validation was not consistently applying the configured tolerance for mean-based outlier rejection. Validation._mean_validation() used the median threshold path, so changing mean_tol had no effect in the scalar mean-validation flow.

• Problem

  • Mean-based GPU validation ignored mean_tol and instead reused median_tol.
  • This made parameterized validation behavior incorrect and hard to verify.

• Code change

  • Updated openpiv/gpu/validation.py so _mean_validation() passes self.mean_tol into _neighbour_validation().

• Regression coverage

  • Added a focused test in openpiv/test/test_gpu_validation_logic.py that loads the GPU validation module with lightweight stubs and asserts mean validation uses mean_tol, not median_tol.
  • The test is CUDA-independent and targets the tolerance wiring directly.

• Behavior before / after

  # before
  _neighbour_validation(..., self.median_tol, ...)
  
  # after
  _neighbour_validation(..., self.mean_tol, ...)