SIMD accelerations

[Nugine]

Motivation
According to my benchmarks, SIMD parser can be 5x~9x faster than generic implementation and SIMD formatter can be 10x faster than generic implementation.

https://github.com/Nugine/uuid-simd/blob/main/benches/main.rs
https://github.com/Nugine/uuid-simd

Solution
Add SIMD functions and a feature simd.
When simd and std are enabled, use runtime cpu feature detection to dispatch the functions.
When simd is enabled, use conditional compilation to dispatch the functions.
Otherwise, use the fallback functions.

Alternatives
SIMD accelerations don't have to be built-in features. An extension crate (uuid-simd) is also good for users who care about performance.

Is it blocking?
No.

Anything else?
Uuid::try_parse is slightly slower than my impl. I'm testing why this happens. I'll submit a PR if I find the reason.

parse/fallback/simple         | 27.286 ns   
parse/fallback/hyphenated     | 28.057 ns
parse/uuid-latest/simple      | 40.266 ns
parse/uuid-latest/hyphenated  | 41.445 ns

Edit:
uuid-simd link: https://docs.rs/uuid-simd/0.5.0/uuid_simd/

[KodrAus]

I think in uuid it would be great to offer something based on std::simd once that becomes available, but the burden of maintaining specialized implementations for different extensions might be a bit much for us. We could add a simd feature that pulled in a library of vectorized parsing and formatting functions to replace uuid's internals with. We should also call out uuid-simd in the README!

[KodrAus]

Since we mention uuid-simd in the README I think we can go ahead and close this one for now. The next opportunity we'll have to action it will be sometime in the future, so we can either re-open or create a fresh issue then.