[RunDevelopment]

Small PR to document which platforms return results with one rounding error for SimdFloat::{mul_add,mul_sub}.

I also changed the parameter names to a and b to match the standard library.

[Shnatsel]

I think "one or two rounding errors" would be more clear.

But I'm not convinced we should document the list of platforms explicitly; if someone needs single-rounding FMA they should use mul_add_precise, not look at the list of platforms and go "well this is all I support so it's fine". Make a portable choice either for guaranteed precision or guaranteed performance.

[Shnatsel]

I think a similar note on performance on mul_add_precise describing where it is or isn't a single native instruction is a better idea.

Admittedly it's the same info, just framed differently.

[RunDevelopment]

No, I like the idea of framing it as a perf note. Will do.