This makes all free functions show up in the generated documentation.

- We now use the same values and formulas as in Smith & Aldrian for the detector 2D variance.
- This changes the ranges of values a bit, so set an equivalent reasonable regularisation default value

Changed the detector variance in the shape fitting from clip-coords to image-coords. This makes the regularisation behave normal again.

Also chose a slightly higher regularisation value to stay a bit closer to the mean face by default.

Added an extract_texture overload that works without supplying a depthbuffer - it should be the default function to use.

- removed a few unnecessary temporary objects
- unnecessary copy-construction of objects
- formatting cleanup (whitespaces)

In the default setting, it links to OpenCV's debug libs, and that doesn't work with the STL of the RelWithDebInfo build.

This increases the readability of extract_texture a lot.

- Added an assert that checks for texcoords
- Pulled over changes from affine renderer - adding const and static_casts instead of C-style casts

- Taking a 3x4 affine matrix and adding an orthogonal row to make it 4x4
- Removed all transformations from clip to screen space as our camera matrix now goes directly from world to screen space
- Removed division by w (it's 1 anyway)

Conflicts:
	CMakeLists.txt: Just merged the new affine renderer headers and texture_extraction.hpp.

- It works with our affine camera matrices estimated from the H&Z algorithm
- Performs depth testing and rasterises triangles
- Vertex-colouring only, no texturing
- Will be used for the texture extraction

Also removed a few trailing whitespaces

Added an extensive check to see if the user is running MSVC, gcc or clang and what version. It's a bit unnecessary to be honest because it's documented. I will possibly remove it after dropping support for gcc-4.8 and -std=c++11.

The assembly on VC14 is the same, but anyway!

For path operator +=