Use Capella fork version for BLSToExecution

[mkalinin]

A stopgap fix for BLSToExecution gossip validation:

• Use Capella fork version If a node is in the pre-Capella state according to a wall clock,
• Otherwise, use a fork version that is relevant.

This fix only make sense if we can't afford a general fix of gossip messages validation because of additional engineering complexity. The general fix would be to always use a fork version from a gossip topic to compute signing domain across all topics.

[dapplion]

I would apply the rule to use the gossip topic's fork version for BLSToExecution topics, even if it's not standard across all topics

[djrtwo]

I'm okay with this but am not too worried about the UX issue if we don't patch. That said, if engineers are comfortable with this, it will ensure we don't get a bunch of dropped messages

[djrtwo]

Just the chain or including any empty transitions to get to the current slot wrt wall time?

[mkalinin]

Good point, I think we don't need the whole state in the definition of this change. I'll switch to system_clock_epoch (Thought this name for the variable would be less ambiguous than current_epoch)

[michaelsproul]

I am in favour of this approach over the partial patch in #3135. I'm also open to the full cleanup of gossip topics, but think we could add that in a future fork to avoid delaying Capella. It doesn't seem like there would be any issue in spec or impl, but the devil is in the details and it would take time from the spec side and every team to roll it out.

[hwwhww]

Per the fork version validation, another solution is making SignedBLSToExecution signature fork-agnostic, i.e., use GENESIS_FORK_VERSION for all SignedBLSToExecution. Although that would make this operation not distinguishable when chain forks.

[mkalinin]

Closing in favour of #3206