Replace boost math functions.

[bitshifter]

Tests added using common input values and comparing against the output from original boost::math implementations.

The normal distribution CDF function is taken from the example code at http://en.cppreference.com/w/cpp/numeric/math/erfc.

The erf_inv is based on code accompanying the article "Approximating the erfinv function" in GPU Computing Gems, Volume 2. Orginally I was using the approximation described at https://en.wikipedia.org/wiki/Error_function#Approximation_with_elementary_functions but found them to be a lot less accurate than the boost::math version for some inputs.

I don't have a great background in this kind of maths, I've found algorithm descriptions through Googling and checked correctness by comparing against the values returned by boost math. It seems to be OK.

[rmartinho]

So this fixes #48? Very cool, thanks! I shall review and merge this with the next batch of free time :D

[bitshifter]

Yep, I think it should fix #48 :)