Sat Mar 9 18:12:44 EST 2013

Exponentials: compute exp(x) and exp(iy) separately?

Maybe not such a bad idea, since both might need separate precision,
and it would free up the dependency chain.

This essentially computes exp,sin,cos.

Using just taylor expansion, stability requirements are easy to
satisfy, as for real exp the approx is strictly smaller, while for
sin/cos it is satisfied if we stop at a negative term and stay in the
first quadrant.