An extended precision floating point library (as per Shewchuk) - precision only limited by overflow / underflow
