Remove build-time check for stdsimd feature

[Amanieu]

This is blocking rust-lang/rust#117372, which replaces stdsimd with more fine-grained features. However the auto-detection in aHash breaks when bootstrapping Rust because it detects the stdsimd feature on the old toolchain which is not present on the newly build libcore.

This PR removes the build-time detection of the stdsimd feature and instead just uses the ARM AES intrinsics directly since they are now stable, but only on AArch64.

[tkaitchuck]

A couple of questions:
This removes support for AES on arm which is not AArch64. Can we leave that in? What will happen there?
When the linked PR is merged, will older versions of aHash fail to build?

[Amanieu]

I'm not a fan of probing for nightly features in build.rs since, even if the feature name stays the same, breaking changes could be introduced at any time. This means that aHash could randomly break on a future nightly compiler. It's better to let users opt-in to using nightly-only features with a nightly feature in Cargo.toml (or even a separate feature for each nightly feature).

I could add an arm-simd Cargo feature which uses the unstable arm AES intrinsics, but it would have to be explicitly enabled by users. It could be removed when the arm neon intrinsics are stabilized. Would that be ok?

When the linked PR is merged, will older versions of aHash fail to build?

No, I think this only happened due to a peculiarity of rustc's bootstraping. The build script was run against the old libcore which used the stdsimd feature, but the new libcore that ahash was actually built with didn't have that feature.

[Amanieu]

I've added back ARM support behind a Cargo feature.

[Amanieu]

Could you publish a new version of the crate?

[Amanieu]

@tkaitchuck Ping! Can you please publish a new release with this change? This is blocking updates of stdarch is the standard library.

[Amanieu]

@tkaitchuck Another friendly ping! Could you please publish a new release?