Experimental offchain state, pt 2

[mitschabaude]

• follow up for Experimental offchain storage pt 1 #1630
• releases the MVP as Experimental.OffchainState
• adds an update() method for which allows a state precondition
  • a single failed precondition causes all state changes in an entire update (method call) to be skipped
  • precondition is in the form of an Option<T> so that we can require the previous state to either exist or not exist
  • field.get() was changed to return an option, otherwise we wouldn't have access to that info
• improve Merkle update efficiency, by first transforming actions into a flat list without dummies due to fewer actions per update
  • previously, 3 updates per batch + 2 actions per update caused 6 in-circuit merkle updates
  • now, 6 actions per batch + n actions per update (n <= 6) cause 6 in-circuit merkle updates
  • the amount of updates per batch we process is no longer uniform. instead, we process a fixed number of actions per batch

TODO:

• make failed updates work (change Merkle update logic everywhere, not just in the offchain prover)