LibTorch: include cmake files for all distributed headers

[Xfel]

I'm planning to add libtorch support to a project that already uses pybind11, and I got a few errors because libtorch also bundled pybind11, which is automatically added to the include path.

While it's easy enough to use the bundled pybind11 in our project, there is no clean way to reference this in cmake since the libtorch distribution does not include a cmake config file for pybind11. I would suggest adding it, so that one can properly reference the bundled pybind11 version.

[soumith]

cc: @goldsborough

[goldsborough]

I guess we need a way to configure the pybind11 install in our CMake. Also cc @anderspapitto @orionr

[famura]

What's the status of this problem?

I have seen that in the latest PyTorch version (1.0.1) a development version of pybind11 is used.