* Improves storage speed by overlapping compression and data transfer
  with computation.
* Changes compression method to local blocks instead of global, to
  improve results (avoid loss of small amplitudes).
* Enables torch.func methods on property models in Python "eager" mode.