Remove restriction that messages are exactly 32 bytes

[jonasnick]

(as discussed here bitcoin-core/secp256k1#995 (comment))

[real-or-random]

Related: sipa#221

[real-or-random]

This has been solved by #24.

But when working on sipa#221, I realized we should have at least test vector with a message larger than 32 bytes. This will catch implementations that truncate the message at 32 bytes.

[jonasnick]

@real-or-random good point. I added test vectors in #52.