normalize_image_gradients(), line, signed_distance_to_line(),
distance_to_line(), reverse(), intersect(), count_points_on_side_of_line(),
count_points_between_lines(), dot(), and normalize() to Python API.

- Moved all uses of PYBIND11_MAKE_OPAQUE to a common header to avoid possilbe ODR violations.
- Added python binding for make_bounding_box_regression_training_data().

* Replace boost::python with pybind11

* Replace add_python_module with pybind11_add_module

* Fix clang error on type-dependent expression

numpy's import_array() in a more robust way.

void*.  But the previous code didn't so it wouldn't compile.  Fixed that
problem.

* Exposed jitter_image in Python and added an example

* Return Numpy array directly

* Require numpy during setup

* Added install of Numpy before builds

* Changed pip install for user only due to security issues.

* Removed malloc

* Made presence of Numpy during compile optional.

* Conflict

* Refactored get_face_chip/get_face_chips to use Numpy as well.

CMake.

Only issue is that the strong is hardcoded - otherwise
this is actually technically required by PEP 396

This aids an API for the correlation based tracker that Davis
recently added to Dlib. I've made sure to allow overrides for
passing normal rectangles rather than drectangles (which isn't
currently supported in the Dlib C++ API). This is mostly
because I imagine people might initialize the tracking using
something like a bounding box from a detector (e.g.
load_frontal_face_detector).

The correlation tracker deals with the drectangle
(double rectangle) class which currently isn't wrapped. Therefore,
I add the drectangle class and refactor rectangles into their
own file. I also added a load of methods on rectangle that might
be useful such as intersection/contains/area etc.

I also cleaned up a bunch of code. I'm not sure why the
simple_object_detector was keeping track of the upsample amount,
since it can't even be passed as an argument to the constructor.
Therefore, I removed the simple_object_detector_py and the second
declaration of the hog object detector. I also changed the
view code to optionally take keyword args of color and added
a single view of a rectangle.

Finally, I added viewing of the shape parts.

This includes the full_object_detection, a new struct in the same
vein as the simple_object_detector_training_options and of
course, the shape predictor classes themselves.

All of training, fitting and testing are wrapped.

Also, move the vectorize template into its own header to
stop having to declare it again in vector.

python.