aead: add inout support

[tarcieri]

Renames AeadInPlaceDetached to AeadInOut, and changes the type signature so the provided buffer is now an InOutBuf

[newpavlov]

I don't think it's worth to gate inout and have the split between AeadInPlace and AeadInOut. If your intent is to use the traits for hardware-based encryptors which can not support buf-to-buf mode of operation, then I think they should implement the inout API by simply copying data from input to output buffer in the case the buffers are separate.

But I guess we can discuss it in separate PRs/issues.

[tarcieri]

If your intent is to use the traits for hardware-based encryptors which can not support buf-to-buf mode of operation, then I think they should implement the inout API by simply copying data from input to output buffer in the case the buffers are separate.

It's for wrapping any API which can't use the inout pattern, whether it's a hardware accelerator, a C implementation of an AEAD, or a TPM/SEP/KMS/etc (which would probably benefit from async traits, but I digress).

I think we can potentially combine AeadCore, Aead, and AeadInPlace into a single trait at this point, but I think it makes sense to keep inout separate.

[baloo]

About gating the inout feature, I don't think there is a single crate that can work without it here: RustCrypto/AEADs#664

[tarcieri]

@baloo as mentioned above, it's gated to ensure the traits are still useful/usable for use cases beyond our own crates

[baloo]

fair enough.