Feature/static dispatch

[calebzulawski]

This is the final result of what I think is the solution for static dispatching. It has a few compromises, from what I can tell.

• The function versions for a safe function are marked safe for each architecture, even though they are not. I don't particularly like this, but it allows static dispatch to inline correctly. They are marked with #[doc(hidden)] and given names that are unlikely to be shadowed, so unless a user seeks out the implementation details I don't think it's likely for this to cause problems.
• The dynamically dispatched function doesn't inline perfectly. This results in an extra jump instruction after the indirect function call. I believe this is due to the functions now being marked safe and the compiler avoiding speculative execution of instruction set extensions.
• Inlining recursion now requires using #[static_dispatch]. Personally I think it's more consistent this way, anyway.

I think these compromises are worth it in exchange for being able to statically dispatch and inline nested functions. @TethysSvensson I'm curious what your opinion is. I'm comfortable with the implementation if you just want to take a look with cargo expand/asm.

[TethysSvensson]

I'm sorry I missed this and didn't reply. I know this is already merged, but I'll take a look know if it's any consolation.

[TethysSvensson]

I again haven't actually looked at how the proc_macro code works, just what it outputs.

I think what it outputs makes a lot of sense. I agree that it is sad that we cannot make the function unsafe, but also agree that it doesn't matter too much in practice.

I do not currently have time to go into detail about what the assembly output looks like, but perhaps I'll come back and give this crate another look at some point.

[calebzulawski]

Thanks for taking a look, I appreciate a second set of eyes!