Consider the following type:

#[derive(IntoBytes)]
#[repr(C, align(8))]
struct Foo<T> {
    t: T,
}

#[derive(IntoBytes)] emits an IntoBytes impl for Foo with a T: Unaligned bound. The reasoning is based on the repr(C) layout algorithm, but this reasoning is unsound in the presence of #[repr(align(8))], which #[derive(IntoBytes)] spuriously ignores.

In particular, Foo<u8> satisfies u8: Unaligned, but has size 8 (7 bytes of padding) in order to satisfy its alignment requirement.

We need to either ban #[repr(align(...))] in #[derive(IntoBytes)] or at least ban it when generics are present.