Previously math-emu was using the IEEE-754 constants internally. These
were differing by having the constants for rounding to +/- infinity
switched, so a conversion was necessary. This would be entirely
avoidable if the MIPS constants were used throughout, so get rid of
the bloat.