Altair: carry-over prev epoch participation

[michaelsproul]

Closes #2314

This is a (draft) PR implementing the translation of phase0 participation information into Altair participation flags at the fork, in order to prevent loss of information (and rewards).

It factors out a function from process_attestation which I had already factored out in Lighthouse anyway for use in the op pool (being able to calculate attestation flag bits independent of processing is nice).

Apart from that it adds a small function called translate_participation, which acts on the pre state's previous_epoch_attestations. I initially thought it should translate current_epoch_attestations but the lists have already been swapped by process_participation_record_updates at the time of the upgrade.

I'll start working on tests shortly. I think the main risk is covering all of the possible attestation behaviours, so I'll look into whether it's possible to reuse generators from the attestation tests.

[hwwhww]

It looks that this solution doesn't increase too much complexity 👍

[hwwhww]

I like this refactoring! IMHO we can apply it even if the translate_participation solution doesn't get included.

[michaelsproul]

Thanks for the review!

[djrtwo]

This was discussed on the latest eth2 implementers call, and there was a decent amount of pushback on the complexity and testing tradeoff.

Looking for more engineering input here before a merge or close

[ajsutton]

This looks pretty straight forward to implement to me and particularly with the refactoring to use the get_attestation_participation_flag_indices method for both this transition and normal operation the additional testing load seems very low. The refactoring probably aids testability. So I don't have a problem with this going in.

I'm not overly concerned about validators being penalised for an epoch - the optics aren't great but it's likely to be forgotten pretty quickly. But given we already have to do a special transition to activate Altair and this is a pretty simple change I'd lean in favour of including it.

[terencechain]

I'm also leaning towards inclusion. Reasons:

• I like the refactoring of get_attestation_participation_flag_indices, that portion is already tested in unit tests and spec tests
• We just need new spec tests on upgrade_to_altair and further practice + verification with this on live testnet

With that said, I'm also not concerned about stakers getting penalized for one epoch, the social implication won't be good, but they will be forgotten rather quickly

[djrtwo]

I'm making a PR to this PR handling the merge conflict and adding tests