Split low-level bindings from high-level abstractions

[michaelsproul]

I think there's a design tension in the library as it currently exists between providing low-level access to GMP primitives and high-level Rustic abstractions.

I think it could be good to split functionality (either internally or as two separate crates), between:

• Unsafe FFI bindings to C functions, with gnarly GMP-style names.
• High-level Rust abstractions with nice names, e.g. Int instead of Mpz, Rational instead of Mpq, etc. No inappropriately low-level methods like Rational::set_f (prefer From, etc).

Happy to help implement this 😄

[vks]

I think the conventional approach would be to have a gmp-sys crate for the low-level functions.

[fizyk20]

Sorry for the delay in replying!
As far as I know, many other crates do what @vks suggests - put the low-level bindings into a *-sys crate and then use this crate in a library of higher-level abstractions. TBH, I haven't seen much sense in that and that's why it's still all in one place, but it's starting to look more useful after your recent PR ;) I'm still not sure what the advantages of a separate crate instead of just a module would be, but separating the low-level bindings somehow definitely looks like a good idea.

[vks]

I see the following advantages of a low-level crate:

1. Others can use it to build competing high-level bindings.
2. It can be used to access functionionality that is not yet exposed by the high-level bindings.