Adjusted all files that use it (hopefully).

This is a breaking change: If you're currently using the PcaModel constructor and giving it a PCA basis matrix, your code will (probably silently) break: The function now expects an orthonormal basis matrix, whereas before, it expected a rescaled basis matrix. However it makes more sense for the function to take an orthonormal basis matrix.

This should clear up confusion and make the descriptions good and unambiguous.

- orthonormal = the old "unnormalised"
- rescaled = the old "normalised".

Also changed PcaModel's getter for the basis accordingly, and updated the documentation.

Both the BFM and SFM store the orthonormal basis in their files. The orthonormal basis is what eos calls "unnormalised" (because it's not rescaled by the eigenvalues).
We will get rid of this ambiguous terminology in eos now and use the unambiguous terms:

- Orthonormal basis (previously "unnormalised")
- Rescaled basis (previously "normalised).

This was in the past used for converting a small BFM json model to a cereal binary. However, it was never satisfactory.
I think it's also used to import the blendshapes from Matlab into C++, so we'll need it again at some point. However at the moment it does not compile as we don't have a json serialize() (cereal) function for Eigen::Matrix.

Also removed division by 1000 of the mean. I don't think it makes sense to make this modification. Leave it as it is in the original.

Top level CMakeLists.txt already check '3.1.3' so there is no need to check
it for subdirectories

'project' declare independent block of CMake code. If block is not designed
to be used stand-alone it make no sense to declare 'project'

If 'REQUIRED' used inside 'find_package' there is no need to check '*_FOUND'
variable. And vice versa - if package is not optional then 'REQUIRED' can be
used instead of 'if(NOT *_FOUND) message(FATAL_ERROR)' block.

Also swapped order of project() declaration and cmake_minimum_required()

clang and gcc complained about this, while VS2015 accepted it. It turns out this is the much better way to do it!

* Now using the transparent glm bindings
* Replaced nonlinear camera estimate with the linear algorithm
* Updated RenderingParameters bindings

Careful about GLM's notation! A 3x4 matrix is a mat4x3!

Based on pybind11/eigen.h, fairly simple.
A bit of care had to be taken for matrices because glm stores the data in col-major layout in memory.
It makes an unnecessary copy of the data in glm::make_mat*, but these types are small anyway.

* Added estimate_orthographic_camera inside the submodule "fitting", and also added bindings for RenderingParameters
* Updated documentation a bit

This allows buffer objects (e.g. arrays) to be converted to eos.Mat (and eos.VecXf), which will then subsequently be converted to cv::Mat/cv::VecXf in C++.
Added mostly support for 1-channel CV_32F and CV_64F types.
The explicit conversion that's required in Python is a bit cumbersome, but I can live with it for now.

Put things from the morphablemodel namespace in a python submodule.

This adds a Python buffer view to cv::Vec4f, and the type is exposed as eos.Vec4f.
That means that the type is convertible to a NumPy array using `np.array(vec4f_instance)`.

Also renamed the app that creates the bindings to generate-python-bindings.cpp.

- Added a flag GENERATE_PYTHON_BINDINGS to CMake which enables the use of pybind11
- `create-python-bindings` generates bindings for PcaModel
- pybind11 is added as a submodule; needs to be cloned with `git submodule update --init`
- Increased cmake_minimum_required to 2.8.12 for pybind11

Added a BUILD_UTILS CMake option, defaulting to OFF.