Consider return messages (and other fields?) as buffers

[davidben]

If you receive a zgram containing invalid UTF-8, we'll decode it into replacement characters. Roost will then encode as UTF-8 before storing into the database. Arguably we want to record the exact bytes in the database, even in the case of invalid UTF-8.

The one interesting thing that will change is that the roost REST API needs to change to pull messages as base64, and then the client decodes. This is somewhat of a pain, so let's only bother with all this if it comes up.

[davidben]

I believe both DES and AES zcrypts already convert the ciphertexts into ASCII. And since messages cannot contain NULs, it is unlikely someone will care about arbitrary bytes in the future.