linalg method of gw loss gives erroneous values with pytorch

[janetrbarclay]

Using the linear algebra calculation approach for the groundwater loss in pytorch gives values 2 to 10 times higher than the loss values using the fft calculation within pytorch, the fft calculation within tf and the linear algebra calculation within tf (these three are the same).

We've added an assertion that requires the fft method when the gw loss is used within pytorch. We are not debugging the linear algebra code, but are leaving it for now in case it is needed in the future.