require scram server nonce to extend the client nonce

[dxbjavid]

When the SCRAM-SHA-256 client processes the server-first message it only checks that the server nonce begins with the client nonce, so a server (or a man in the middle) can echo the client nonce unchanged and the client carries on as though the handshake were sound. RFC 5802 section 5.1 requires the server to append its own nonce to the client's, and accepting an unchanged value drops the server's contribution to the combined nonce and the freshness that comes with it. This tightens the check in ScramScheme to also require the server nonce to be strictly longer than the client nonce, with a regression test; conformant servers always extend the nonce, so interoperability is unaffected.