Partially addresses #1709 and related to the old #1667 that was closed to address each issue independently.

After Matthew's input here, we now know that MIPS and Keccak columns won't intermingle because we will be having two different instances (one for each). This means that we don't need to consider the case that a row is not a sponge nor a round, but an instruction. Consequently, we can be fine with a negated definition of is_round() as not(is_sponge()), reducing by one not only the columns (because we no longer need inverse_round to check that the round number is nonzero) but it also removes one constraint. It removes the dummy entry of the round constants table, as the 0 entry now corresponds to the first keccak round (as it used to be before #1596)

This simplifies the code and reduces some code duplications in the main.