crypto-common: potential future of `KeyIvInit` for `T: InnerInit` where `T::Inner: KeyIvInit`

Moving this commented out code from the crypto-common source to a tracking issue:

Unfortunately this blanket impl is impossible without mutually exclusive traits, see:
https://github.com/rust-lang/rfcs/issues/1053

or at the very least without:
https://github.com/rust-lang/rust/issues/20400

impl<T> KeyIvInit for T
where
    T: InnerInit,
    T::Inner: KeyIvInit,
{
    #[inline]
    fn new(key: &Key<Self>, iv: &Iv<Self>) -> Self {
        Self::inner_init(T::Inner::new(key, iv))
    }

    #[inline]
    fn new_from_slices(key: &[u8], iv: &[u8]) -> Result<Self, InvalidLength> {
        T::Inner::new_from_slice(key)
            .map_err(|_| InvalidLength)
            .map(Self::inner_init)
    }
}

Start with the commented-out KeyIvInit implementation in the crypto-common source and review the linked Rust RFC and issue about mutually exclusive traits and specialization. Determine whether current Rust trait capabilities permit this blanket implementation; done requires a documented, feasible resolution rather than merely restoring the commented code.