Make checkpoint sync work when finalized state is transitioned with empty slot(s)

[StefanBratanov]

PR Description

Override the consistency check for an AnchorPoint since the finalized state can be transitioned with empty slot(s)

Fixed Issue(s)

fixes #7956

Documentation

• I thought about documentation and added the doc-change-required label to this PR if updates are required.

Changelog

• I thought about adding a changelog entry, and added one if I deemed necessary.

[zilm13]

I like the way you've kept the check in Summary while avoiding it in some cases in AnchorPoint but I still have a lot of doubts on 2 major points of this change:

• Could weakening of AnchorPoint creation checks introduce a bugs when AnchorPoint is created not from checkpoint sync it has not failed, but it's a bad AnchorPoint
• How transitioned AnchorPoint could affect its consumers when we start Teku

While from quick check I don't see any issues with 1st, 2nd could be definitely an issue in many places. I've commented one case in the Store, another one for example is initalization of the Store:

    // Track latest finalized block
    this.finalizedAnchor = finalizedAnchor;
    this.maybeEpochStates = maybeEpochStates;
    states.cache(finalizedAnchor.getRoot(), finalizedAnchor);

This way we are putting incorrect cache in states cache which is widely used. And there are a lot of other places where finalizedAnchor spreads, we should check them all and decide how to handle it. Maybe this kind of Anchor will need extra identification so we could avoid polluting some things like this state cache with it.

[zilm13]

I'd also add verification for state.block_roots (block slot - state slot interval) and state.state_roots (block slot)

[StefanBratanov]

Added those checks, made it bit more complex, since don't have access to spec in super class.

[zilm13]

If it's transitioned anchor, it will be incorrect state for requested block_root

[StefanBratanov]

AnchorPoint.getStateRoot() still returns the state_root of the finalzied block (last block header), not the state

[zilm13]

Method private SafeFuture<Optional<StateAndBlockSummary>> getOrRegenerateBlockAndState( final Bytes32 blockRoot) should return state at block with blockRoot, transitioned state is not this state. This method is used by a lot of different consumers, it could be an issue

[StefanBratanov]

Yeah AnchorPoint.getStateRoot() returns the state_root of the blockSummary, so would be the state associated with the block

[StefanBratanov]

In any case, StateAndBlockSummary.create(finalizedAnchor.getBlockSummary(), finalizedAnchor.getState()) is same as just returning finalizedAnchor

[tbenr]

On startup we definitely create the Store with the anchor coming from the checkpoint URL. So it is possible that we could return it from here...

[StefanBratanov]

Closing this in favour of #8221