Intermediate state roots about celestia-core

[adlerjohn]

Based on the leaps and bounds progress in the ZK space, I'm heavily leaning towards just going directly to validity proofs and skip fraud proofs for the state machine.

[liamsi]

This can be split up into the following steps (of which some can and should be worked on in parallel):

1. make the preprocess step for block N return the execution trace of block N-1 and include it (the interm. state roots) in a block
2. remove the validator roots from the header and fold these into state root #66 (this will break the light client that relies on these roots to exist but we'll modify the light client to understand and use the relevant subtree of the SMT used in the abci app)
3. Introduce an abci method FinalizeBlock and another one (before prevoting) for the non-proposer nodes to also trigger (and validate) the pending state transitions in the app, FinalizeBlock is needed to actually commit the pending new state to disk (app and tendermint side)
4. make the preprocess step for block N return the execution trace of block N and include the state roots in a block, also update the single state root when preprocessing - this will enable a single state fraud proof

[evan-forbes]

are fraud proofs still the current plan or are we leaning towards just going straight to using a snark to prove the state machine? @adlerjohn @liamsi @musalbas