Improved accuracy of libm

[ZERICO2005]

Changes:

• added header_test.c and header_test.cpp to test for conflicts
• documented the ULP or relative precision of functions
• added 3-argument hypot to C++, it can also be used in C through <tgmath.h>
• implemented expl logl atanl atan2l asinhl acoshl atanhl erfl erfcl
• optimized various float32 math functions
• __float32_constants.h and __float64_constants.h were added to maintain consistency.
• Added basic tests to for each math function (tests a positive and negative input, and sees if the result is "accurate" enough (within 512ulp))