Removed reliance on nightly

[oh-yes-0-fps]

This is a rough pass on using wide instead of portable_simd for vectorized operations.

It passes all the tests and seems comparable on benchmarks

Portable Simd

Wide

Full disclosure the first pass of this was done using claude code with bug fixes and regressions being tracked down manually.

[claytonwramsey]

Thanks for the PR!

I don't mind moving off of nightly, but I think I can't accept this PR for two reasons:

1. I have some yet-unpublished projects that also rely on std::simd, and don't have the bandwidth to port everything.
2. I want to be able to support arbitrary lane selections and floating-point data types -- e.g. for AVX-512.

Ideally, if we pick a third-party SIMD library, it should also support scatters and gathers (though on most hardware those ops are effectively sequential).

[oh-yes-0-fps]

I understand. You could use some basic conditional compiling to implement your own gather.