bls: Use int_to_bytes8 to serialize domain

[wemeetagain]

The function bytes8 does not exist in the spec. Using int_to_bytes8
has a clearer meaning: serialize into 8 bytes as little endian.

[hwwhww]

1. IMO we would avoid BLS spec referencing to core specs.
2. int_to_bytes8 is little endian converting, while x_re and x_im are still in big-endian. IMO BLS spec is part of cryptography black box (Little vs big endian #556 (comment)). Also, related to Cross-blockchain BLS12-381 standardisation #605.

What do you think about explicitly:

    x_re = int.from_bytes(hash(message_hash + domain.to_bytes(8, 'big') + b'\x01'), 'big')
    x_im = int.from_bytes(hash(message_hash + domain.to_bytes(8, 'big') + b'\x02'), 'big')

[wemeetagain]

If we clarify as you mentioned, that gets at the heart of the issue, that the endian-ness is currently unspecified/unclear.

Perhaps an even cleaner solution would be to pass in domain as a bytes8 rather than as a uint64, and have the bls spec not need to worry about domain integer serialization at all.

eg:

def hash_to_G2(message_hash: Bytes32, domain: Bytes8) -> [uint384]:
    x_re = int.from_bytes(hash(message_hash + domain + b'\x01'), 'big')
    x_im = int.from_bytes(hash(message_hash + domain + b'\x02'), 'big')

[hwwhww]

Hey @wemeetagain, I like your second proposal more. :)

However, it will change the interface, and #605 may change it again later. Therefore, it may not worth it to change the interface recently before #605 due to the overhead of updating test vectors.

[wemeetagain]

Oky, I'll close this PR. When the #605 change rolls around, we can give the appropriate input.