Update deterministic subnets upgrade to allow prefix computation

[divagant-martian]

Issue Addressed

Implements ethereum/consensus-specs#3545 in preparation for turning on deterministic subnet peers discovery

Proposed Changes

Add the spec values necessary to implement the change so that peer discovery can be tackled afterwards

Additional Info

found the 🐛 and now all the test values are verified with pyspec so I'm certain calculations are correct

The compute_subscribed_subnet function required two changes to work:

1. proper types for subnet_prefix and shuffling_bits
2. using parenthesis to specify the correct order of operations in python for the later (otherwise the result might not even fit in a uint64).

For the record this is the code:

def compute_subscribed_subnet(node_id: NodeID, epoch: Epoch, index: int) -> SubnetID:
    subnet_prefix = uint64(node_id >> (NODE_ID_BITS - config.ATTESTATION_SUBNET_PREFIX_BITS))
    shuffling_bits = uint64(node_id >> (NODE_ID_BITS - config.ATTESTATION_SUBNET_PREFIX_BITS - ATTESTATION_SUBNET_SHUFFLING_PREFIX_BITS) )% (1 << ATTESTATION_SUBNET_SHUFFLING_PREFIX_BITS)
    epoch_transition = (subnet_prefix +  (shuffling_bits * (config.EPOCHS_PER_SUBNET_SUBSCRIPTION >> ATTESTATION_SUBNET_SHUFFLING_PREFIX_BITS))) % config.EPOCHS_PER_SUBNET_SUBSCRIPTION
    permutation_seed = hash(uint_to_bytes(uint64((epoch + epoch_transition) // config.EPOCHS_PER_SUBNET_SUBSCRIPTION)))
    permutated_prefix = compute_shuffled_index(
        subnet_prefix,
        1 << config.ATTESTATION_SUBNET_PREFIX_BITS,
        permutation_seed,
    )
    return SubnetID((permutated_prefix + index) % config.ATTESTATION_SUBNET_COUNT)

[divagant-martian]

Tagging @michaelsproul since you have good eye for these things

[AgeManning]

These changes look fine to me.

The conversion to u64's I think are fine, but if someone wants to double check it can't hurt us that might be one area to look at.

There is edge cases for panics, but only on wild configuration settings, so i'm cool with that.

[divagant-martian]

Yes, I am assuming that the number of bits we take for subnet prefix + shuffling will never reach 64, since this would require us to store a mapping for 1.8×10¹⁹ (2^64) node id prefixes -> subnet, per epoch of interest (probably at least the current and next one).

Note that the previous code has similar assumptions, since node_id_prefix had to fit in a usize so we assume that the prefix bits are also less than 64.

I don't think these assumptions are outrageous, since they are controlled by the spec and anyone changing them to crazy values can't cause harm to other nodes but their own.

But ofc happy to take suggestions from @michaelsproul when he gets the time

[dapplion]

Merged unstable to fix some conflicts with Dencun variables.

Logic looks good and matches ethereum/consensus-specs#3545

Should add a check at the start compute_subnets_for_epoch to check that prefix_bits + shuffling_prefix_bits < 64?

[AgeManning]

I think we get this one in. @ackintosh has another that is dependent on this one also.

[michaelsproul]

@AgeManning gentle reminder to use mergify next time 🙏

@Mergifyio queue is the new bors r+