Add rounding and casting functions to math::fxp

[jakobj]

This PR introduces rounding to the fixedpoint type (currently only used in mul) and functions to up-/down-cast to other fixedpoint types with different bitwidths. I've also fixed some whitespace issues.

@rmanohar currently this requires the user to define a global variable FXP_ROUNDING_MODE when using the fixedpoint library, e.g., via commandline flags. not sure this is the best solution, but it works for me.

[rmanohar]

The global is a bit problematic, since a normal user that imports math will just get an error about a missing global. One option would be to add it as a template parameter with a default value. We probably should have a discussion about other options for this.

[jakobj]

right, i thought they'd only get the error when importing fxp, so this solution is suboptimal (unless we remove fxp from __all__).
i agree a template parameter would be nicer (might even the best option?). how would that work with the fixpoint type? my naive approach to using template<> on methods failed.