We obviously have std::mem::transmute, but for core we only have
core::intrinsics::transmute: unstable / feature gated
core::mem::transmute_copy: which is actually more dangerous than transmute because it's missing the "before and after have the same size" check.
So can we get core::mem::transmute?