fix z-coeffs skip when 0 for torch

[edudc]

This fixes torch autograd for trainable Zernike coefficients initialized at zero.
The current Zernike loop skips any coefficient where coefficient == 0. For a torch tensor/Parameter it removes the coefficient from the computation graph.

Proposed change:

  if (n - abs(m)) % 2:
      continue
  if not (TORCH_AVAILABLE and torch.is_tensor(coefficient)) and coefficient == 0:
      continue