Fix calculation of divergence_loss()

[aschuh-hf]

The divergence loss as implemented takes the sum of all first order derivatives.

deepali/src/deepali/losses/functional.py

Lines 1356 to 1358 in ea3c16d

	derivs = spatial_derivatives(u, mode=mode, order=1, sigma=sigma, spacing=spacing)
	derivs = torch.cat([deriv.unsqueeze(-1) for deriv in derivs.values()], dim=-1)
	loss = derivs.sum(dim=-1)

Instead, only the trace of the Jacobian matrix should be computed.