Use two part doubles to improve precision

[matzipan]

In #4 I implemented the ephemeris epoch with an f64. Unfortunately, due to the dynamic range limitations of floats, if the magnitude of the integer part of the float is too high, this will lead to a loss of accuracy.

The way to improve on this is to use two-part floating point numbers, where one holds the integer part and the other holds the fractional part.

This issue is about improving the SPK ephemeris code to use two part doubles.