Shall we split the `std::simd` module into a crate separate from std?

[gnzlbg]

I have a pretty much equivalent implementation of stdsimd that follows the RFC in https://github.com/gnzlbg/ppv

Given that:

• the stabilization path for std::simd is not clear yet, and that stabilization might take a while
• there are issues with some targets failing with core upstream, e.g., due to linkage issues
• std::simd costs most of the CI time of the stdsimd repo,

I'd like to discuss whether we should split std::simd into a crate separate from std (not necessarily mine, we can put the current implementation 1:1 as is in the nursery).

Splitting the crate might let us sync upstream more often and without breakage with stdsimd, significantly reducing the amount of churn we are seeing.

cc @alexcrichton

---

This basically turns rust-lang-nursery/stdsimd into stdarch, and the current std::simd into a crates on crates.io that people can just use with nightly in an opt-in way.

[alexcrichton]

Seems reasonable to me!

[hdevalence]

Hi, I was using core::simd. Would it be possible to not remove it until the replacement is available on crates.io?