document differences from pure hardware behavior

[RalfJung]

Fixes rust-lang/rust#153990 by documenting that the NaN rules also apply to stdarch intrinsics.

Rust does not have a general rule of "signaling NaN may be treated like quiet NaN"; that only affects a few operations (pow, powi, max, min) which just then state this caveat in their respective docs. My understanding is that stdarch wants to properly reflect hardware behavior here. That means however stdarch must be careful not to use the normal LLMV intrinsics for such operations. I think nvptx currently may be getting this wrong, which is already tracked at #2056.

@Amanieu also mentioned that for subnormals, the intent is that stdarch matches the hardware. I suspect this may be tricky since LLVM will probably reserve the right to perform constant propagation with full subnormal precision, i.e., "guaranteed subnormal flushing" is probably not going to work. But right now the ARM32 NEON intrinsics are unsound and can cause UB (rust-lang/rust#129880) so we can't really even have that discussion yet until that is fixed.

[rustbot]

r? @sayantn

rustbot has assigned @sayantn.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: @Amanieu, @adamgemmell, @davidtwco, @folkertdev, @sayantn
• @Amanieu, @adamgemmell, @davidtwco, @folkertdev, @sayantn expanded to Amanieu, adamgemmell, davidtwco, folkertdev, sayantn
• Random selection from Amanieu, adamgemmell, davidtwco, folkertdev, sayantn