Possible inconsistency in the API

This was discussed briefly on irc.

The problem is that when using encoders for signing and verifying pynacl also uses the encoding in the message being signed or verified. Example:

> verifey_key.verify(b'Hello', signing_key.sign(b'Hello', encoder=nacl.encoding.HexEncoder).signature)
> BadSignatureError

My proposal is for pynacl not to assume anything about the encoding of the message and only use the encoding on the signature.

When signing:

signature = encoder.encode(raw_signed[:crypto_sign_BYTES])
message = raw_signed[crypto_sign_BYTES:]
signed = encoder.encode(raw_signed)

And when verifying:

if signature is not None:
    smessage = encoder.decode(signature) + smessage
else:
    smessage = encoder.decode(smessage)

I wouldn't mind creating a pull request for these changes.

Start from PyNaCl's signing and verification API, focusing on the encoder handling shown in the examples. Check how signing and verification currently apply encoder.encode and encoder.decode to the message and signature; done means encoded signatures verify correctly without requiring the message itself to use that encoding.