Input is always parsed as f64s, which can result in precision errors

[IntrepidPig]

This is most noticeable when inputting decimals, for example:

eval::<ComplexRugRat>("2.2") // = becomes 2476979795053773/1125899906842624

However, it also happens when integers become large enough

eval::<ComplexRugRat>("200000000000000000000000") // = 199999999999999983222784

This can be fixed by parsing the number (in parse.rs) with the functions provided by rug. However this will require adding parse functions to the Num trait and passing the N generic type all the way down which will be quite a bit of work.