* Replace boost::python with pybind11

* Replace add_python_module with pybind11_add_module

* Fix clang error on type-dependent expression

any object part labels so that the flipped dataset has the same average part
layout as the source dataset.  I added a --flip-basic that behaves like the old
--flip.  However, most people flipping a dataset with part annotations will
want to use --flip.

appear to support cuda.

Set `CUDA_PROPAGATE_HOST_FLAGS=OFF` in FindCUDA.cmake to avoid duplicate flag errors in nvcc that can occur when host flag forwarding is enabled and `-std=c++11` is set in `CMAKE_CXX_FLAGS` in the operative toolchain.
```
# Override the FindCUDA.cmake setting to avoid duplication of host flags if using a toolchain:
option(CUDA_PROPAGATE_HOST_FLAGS "Propage C/CXX_FLAGS and friends to the host compiler via -Xcompile" OFF)
```

iOS apparently.

be more future proof and in particular correctly enable 64bit compiles in visual studio 2017
when appropriate.

when building dlib in visual studio.

On systems such as FreeBSD, `uname -p` returns amd64 (in lowercase) and -DUSE_AVX_INSTRUCTIONS=1 does not work (the option is ignored).