Tensor Storage

TorchLean's storage boundary selects the physical tensor representation. Arbitrary element types use ordinary Array storage, while types with a specialized instance, such as Float and UInt8, use packed native buffers.

Storage is inferred. Ordinary users continue to write Tensor Float [2, 3] or Tensor Real [2, 3]; generic library declarations bind [Storage α] so specializing α also specializes its storage.