Update aarch64 to use neon as fp

[workingjubilee]

As discussed in rust-lang/rust#91608
At the very least, which feature to use for assembly should be clear.

[workingjubilee]

@ehuss I know that the reference is considered non-normative at the time, but we might want to strike the entire "implications" column for the AArch64 suites of features, and possibly x86-64 as well, as they depend mostly on a property of LLVM and not of Rust, and it's not clear how exactly they match the "true" implication logic (after doing some review, I believe they follow all the correct rules but are heavily incomplete). We would probably want to restore them, but mostly after figuring out what's actually correct and maybe altering Rust's target feature logic also. What say you?

[ehuss]

I don't object to removing the column. I think my concern was the situation where someone has #[target_feature(enable = "foo")] and they rely on it also enabling the bar feature. If in some hypothetical future that implicit feature is removed (like in a different backend), that could break the author's expectations.

I agree it would be better for rustc to be a little more explicit about how these are enabled. There also hasn't been much discussion about this behavior, and it isn't entirely clear to me if it is even all that important.

[ehuss]

Thanks! I'm going to merge assuming I understand rust-lang/rust#91608 correctly.