Functional

TorchLean functional helpers in the style of torch.* building blocks.

These are derived ops built from the small primitive TorchLean.Ops surface, so eager execution and compiled graph construction share the same model and loss definitions.

For background, see the PyTorch documentation for torch.nn.functional, torch.autograd, and checkpointing, together with the standard reverse-mode AD references by Linnainmaa and by Griewank and Walther.