The SipHash algorithm starts with initializing the 32 bytes of internal
state with some magic numbers XORed with the hash key. However, our
implementation has a bug - rather than XORing the hash key, it *sets* the
initial state to copies of the key.
I don't know if that affects any of the cryptographic properties of SipHash
but it's not what we should be doing. Fix it.
Signed-off-by: David Gibson