aead: API with separate buffers for input and output

Would it make sense to add the following methods to AeadInPlace (similarly for AeadMutInPlace) with default implementations based on the existing methods?

fn encrypt_mut(&self, nonce: &Nonce, aad: &[u8], plain: &[u8], cipher_tag: &mut [u8]) -> Result<()> { ... }
fn encrypt_mut_detached(&self, nonce: &Nonce, aad: &[u8], plain: &[u8], cipher: &mut [u8]) -> Result<Tag> { ... }

fn decrypt_mut(&self, nonce: &Nonce, aad: &[u8], cipher_tag: &[u8], plain: &mut [u8]) -> Result<()> { ... }
fn decrypt_mut_detached(&self, nonce: &Nonce, aad: &[u8], cipher: &[u8], tag: &Tag, plain: &mut [u8]) -> Result<()> { ... }

This would add support (without extra copies or allocation) for implementations that don't support aliasing input and output. They would implement the _mut version and use a copy to implement the _in_place version using the _mut version.

Start by reading the AeadInPlace and AeadMutInPlace traits and their existing in-place methods. Determine whether the proposed separate input and output methods fit both traits and can use default implementations; done means the API supports non-aliasing implementations without extra copies or allocation.