Motivation for this change

libtorch is available as python3Packages.pytorch.{out,dev}. However, I
think there are several good reasons for also providing upstream's
prebuilt libtorch:

• Upstream has a license to distribute PyTorch with MKL. I benchmarked
  one of my own natural language processing tools and this resulted in
  50% faster CPU inference. We support building with MKL in our
  PyTorch derivation, but that incurs a large build time.
• Upstream also has a license to distribute CUDA with MKL. Again, we
  support CUDA in our derivation, but it requires an expensive PyTorch
  build.
• No build time. This is nice for downstream users when the PyTorch
  derivation is temporarily broken, which is typically only found out
  somewhere during the build.
• Possibly breaks less often than builds of the PyTorch derivation.

Since the same libraries are distributed through PyPi, I would assume
that redistribution is fine. But I can ask upstream if necessary.

Things done

• Tested using sandboxing (nix.useSandbox on NixOS, or option sandbox in nix.conf on non-NixOS linux)
• Built on platform(s)
  • NixOS
  • macOS
  • other Linux distributions
• Tested via one or more NixOS test(s) if existing and applicable for the change (look inside nixos/tests)
• Tested compilation of all pkgs that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review wip"
• Tested execution of all binary files (usually in ./result/bin/)
• Determined the impact on package closure size (by running nix path-info -S before and after)
• Ensured that relevant documentation is up to date
• Fits CONTRIBUTING.md.