use `rug::Float` instead of hard-coding `f64` for arbitrary precision

[alessandromazza98]

Hi,

have you ever thought about using rug::Float type instead of hard coding with f32 or f64 in order to let people that use your crate to have arbitrary precision over the data that are fed into the model.

I think it would be beneficial not only for me, but for all use cases that require a better precision than what an f64 type can provide - an f64 has 53 binary digits precision.

I would be open to work on the integration of Float instead of f32 / f64 if you are open to discuss it / merge it later.

What do you think?

Thanks, Alessandro