Special function support for interpreted floating point

[robdockins]

In order to represent programs that use trigonometric and complex mathematics, we should add representation support for special functions on real and floating-point values. Of course, very little solver support for such things exists, but we can at least treat such functions as uninterpreted, or perhaps state useful range restrictions or identities about them.

This will be necessary to, e.g., represent most of the standard C functions from math.h and similar constructs in other languages.

[robdockins]

At a minimum, we should support standard and hyperbolic trig functions, exponential and log functions. We might also look at things like gamma, erf, Bessel functions, etc.

[robdockins]

PR #160 makes strides in this direction. I'm not sure when we should call this "done" but this initial work adds support for what i think are the most popular functions.

Currently, the SpecialFunctions module exposes information about domain, range and symmetry properties of these functions that are not used anywhere. Moreover, the IEEE-754 floating point mode and the real mode do not handle any of the special functions. Probably, we should add, at a minimum, support for treating these functions as uninterpreted.

[RyanGlScott]

#163 adds support for Pow.