Replace `memory` with `base{16,64}`

[endgame]

We should aim to stop directly depending on memory: Vincent's packages are unmaintained, there's no clear replacement like with crypton/ite, and it doesn't seem like we're using it for anything fancy. A quick grep makes me think it's only used to do base16 and base64 encoding, so replacing those functions with ones from packages base16 and base64 might be easy.

[endgame]

This is not straightforward: the various Digest as that we get from crypton can only be interacted with using the ByteArrayAccess interface from memory, which is why amazonka-core has polymorphic functions for base{16,64} encoding. And if we move to botan we'll be able to get encoders from there anyway, so there's little point in doing this refactoring twice.

[mbj]

I think we'll have to consider to raise this question with the crypton maintainers, I'd hope they can maintain memory to serve crypton?

[endgame]

That may be necessary in the medium-term, but memory also seems to be small enough that it doesn't seem to need too much active maintenance. Especially not in the amazonka-core use case, where we're just computing digests and HMACs but don't need the crypto primitive stuff like scrubbed memory, constant-time equality, etc. My hope is that we can eventually move all of this stuff across to botan when it matures.

[mbj]

@endgame I think that http-client-tls is used by amazonka, and this uses the tls package which uses crypton which uses memory.

[endgame]

Almost certainly. The migration will be an effort from all of us, but when botan is ready we can at least do our part in the amazonka space.