tensorplay.library

Choosing the kind of custom op

`tensorplay.library.custom_op`	Define a user operator; use as a decorator (torch parity).
`tensorplay.library.triton_op`	Define a Triton-backed operator (`torch.library.triton_op`).
`tensorplay.library.wrap_triton`	Mark a Triton kernel as launchable from within a `triton_op`.

`tensorplay.library.register_kernel`	Top-level kernel registration (`torch.library.register_kernel`).
`tensorplay.library.register_autograd`	Top-level autograd registration (`torch.library.register_autograd`).
`tensorplay.library.register_fake`	Top-level fake-kernel registration (`torch.library.register_fake`).

Scoped registration bundle (torch.library.Library subset).