Generalize encoding and decoding to work for all numbers that implement num_traits::Float

[Trevader24135]

Having the encoder work generally with either f32 or f64 is quite helpful to me. I like to run simulations with higher resolutions using f64, but if I'm creating a piece of software that should keep up in real time, I prefer to use f32. Being able to use either is nice.

This pull also adds criterion benchmarks so that if I find the time, I can try to optimize some of the algorithms used in this library to speed up it's performance, because I find that it's slower than other implementations such as Aff3ct.

Finally, this pull also separates out the dependencies required for the binary crate this library provides into their own feature which is disabled by default, so that consuming projects don't pull in unnecessary dependencies. This perhaps should have been it's own pull request, but I went ahead and did it anyways. If that's undesirable, then I can go back and undo that for now, though long-term I personally think it's a good idea.

[venkat0791]

Thanks, @Trevader24135. Kindly note that it will take me a while to get to this.