Allow the first argument of #[simd] functions to be f32x4<S> and such

[Shnatsel]

Allow the first argument in #[simd] functions to be anything simd-token-carrying, not just a literal _: Simd

Heavily based on the ideas @RunDevelopment outlined in #379

Closes #379

I want to get it in before v1.0 because this is a small semver-breaking change (method moved from one trait to another)

@RunDevelopment please take a look

[Dr-Emann]

I know it would be quite the huge change, but the naming feels inconsistent throughout between Simd vs token vs witness. Is there a distinction I'm missing, or could we call them all the same thing? (Simd, ExtractSimd::simd)

[Shnatsel]

I think Simd would have been SimdLevel if you didn't have to keep repeating it on every function.

I'm open to ExtractToken::token to make the naming more consistent.

[Dr-Emann]

I think Simd would have been SimdLevel if you didn't have to keep repeating it on every function.

Or maybe SimdToken, based on this pr.

I still think vector.simd() better matches, especially with vector.simd already existing. vector.token() is less obvious to me (but better than vector.witness())

[Shnatsel]

Renamed. Any concerns other than naming?

[Dr-Emann]

I still say .simd() would be better

[Dr-Emann]

Either way, definitely a breaking change to note for 1.0

[Shnatsel]

This PR adds a mention to the changelog: https://github.com/linebender/fearless_simd/pull/383/changes#diff-06572a96a58dc510037d5efa622f9bec8519bc1beab13c9f251e97e657a9d4ed

[Dr-Emann]

Saw the added section, missed the changed section somehow, awesome!