Implement EndBlock execution state

[ed255]

Currently we have an EndBlock gadget that is dummy (it doesn't have any constraints).

The EndBlock state was specified here privacy-scaling-explorations/zkevm-specs#88 , nevertheless there are a few details that may be missing from the spec:

# TODO: Introduce constrain_instance to constrain the equality between witness
#       and public input, for total_tx and total_rw

• Constraint that the number of tx processed in the EVM is the same as the number of txs in the tx table
• Constraint that the number of ops in the rw_table from the State circuit match the number of ops "emitted" from the EVM circuit.

Dependency:

• #711

[han0110]

Since currently we have variable step height for each ExecutionState, using EndBlock to pad all remaining rows might not always work because it's not single row, it could possibly go into blinding factors regions (zk) or wrap around (non-zk) and break some constraint.

One approach to make it EndBlock single row (and repeating EndBlock until last enabled q_step) is to turn off zk and gain one extra degree to make ExecutionState selectors fewer, and at the same time not make the circuit wider.

Another very naive approach that just comes to my mind is to introduce a new fixed column global_offset, which is filled with the index of row it's is in (as 0..2k - 1), and in EndBlock we constrain the next row's num_rows_until_next_step to be 2k - (1 + global_offset) (might need to minus blinding_factors() if zk), and q_step will never be 1 again.

Also it seems we don't need to share public input anymore (so no need of constrain_instance) thanks to the idea introduced by @ z2trillion in #558 (for checking number of tx, we might be able to also adapt the same trick).

[aguzmant103]

Sub-tasks:

• #711

[aguzmant103]

Still in progress, missing one review.