Cannot replay epoch info generation for epoch height 1055

[Longarithm]

Epoch height 1055 is around the last height with protocol version 48 used. Apparently what matters is that we started to use ProtocolFeature::AliasValidatorSelectionAlgorithm after that. Looks like we can't replay the old behaviour, attempt to do so results in the first validator index diverging.

It is not visible and has moderate urgency, because everyone seems to use already generated epoch infos, but if one tries to regenerate epoch infos from genesis, their chain is likely to diverge.

Consider

• reusing 1.21.0 to see where the difference comes from;
• use #11343 to write a test which freezes behaviour for all protocol versions.

[wacban]

Is this something that needs to be solved for epoch sync?

[Longarithm]

Yeah, looks like the doc includes recomputation of epoch infos. cc @posvyatokum