Currently the serialised forms and the form fed into the contract aren't quite joined up, because on disk/network a transaction is serialised with input references i.e. (txhash, output index) pairs, but the contract wants to see all input states in fully loaded form. To do that, we need some notion of a database of transactions.
Introduce three ways to represent transactions: wire, ledger and for-verification.
TransactionForVerification contains the arguments formally passed to verify(). Making Contract.verify() take only a single object as the argument makes it easier to evolve the Contract binary interface without breaking backwards compatibility with existing contracts that are uploaded to the ledger. The Kotlin "with" construct can be used to bring the members into scope, thus acting syntactically like function arguments.
Make the contracts to be run depend on both input AND output states: this approach seems simpler than trying to have some notion of marking outputs that are checked. The TransactionForVerification class implements this simple logic (it's all moved into Transactions.kt).
You can now take deposits from multiple different institutions and move/combine/split them appropriately. The issuers are kept separate, you cannot merge 3 different input states from 3 different institutions down to one, but you can merge/split within that specific issuer. Deposit refs are not currently being kept separate, but they should be also (this is coming next).
