…, #128 & #164
This removes the last trace of NSDecimalNumber, which has some surprising behaviour. It allows more precision but less range than double. This required:
* Removing tests for real numbers that cannot be represented accurately using IEEE floating point.
* Avoid rounding issues in tests when all we want is to make sure exponent is handled.
For clarity this also renamed tests for (positive and negative) zero for real numbers.