Docs on std::mem::transmute incorrectly claim the types must have the same alignment

[lilyball]

The std::mem::transmute docs says:

Both types must have the same size and alignment, and this guarantee is enforced at compile-time.

Experimentally, this function does not care about alignment:

#![feature(core)]
use std::simd::*;
use std::mem;

fn main() {
    let _ = unsafe { mem::transmute::<_,(u64,u64)>(u64x2(1,2)) }; // compiles just fine
}