[ENH] Add torch engine for BiphasicAxonMapModel

[jgranley]

Support engine='torch' for BiphasicAxonMapModel. The design should match as closely as possible the code for torch engine for #611 (in progress), meaning it should have a separate class that inherits from torches nn.Module, and predict_spatial is refactored to direct calls to the torch module if engine is torch. The torch module should have a function that takes in the stimulus, model parameters, and electrode locations (as torch tensors) and should return the percept as a torch tensor. All existing unit tests should pass.

If it is convenient then this could use the same torch module as AxonMapModel.