Add tests for payday behavior when balance is negative

[chadwhitacre]

In limited situations, we send a participant's Gratipay escrow balance negative. For example, when they receive stolen money that we refund (https://github.com/gratipay/violations/issues/34), or when I forget to post back MassPay and subsequently overpay people (gratipay/inside.gratipay.com#951). In these situations, we want to recoup the loss from future takes, but our current behavior is to charge their credit card (if they have one) to make up the difference. That penalizes the participant for a mistake on our part, so I think the best thing to do is update payday to not charge a card when a pre-payday balance is negative.

Update: We ended up working around the charges for the 10 people in question by manually overriding the error field on their route to prevent charging. The tests on this PR are still valuable as documentation of existing behavior.

[chadwhitacre]

I think that has implications for people who would've received from these people, though.

A → B
B → C

We double-pay B and send their balance negative to correct. Next week they receive from A again, but that just brings them up to 0 and they don't have enough to pay C.

Sorry, C. :-(

[chadwhitacre]

Or can we be smart enough to charge if the net for the payday is positive? Seems like we should be able to.

[chadwhitacre]

This is complicated because new_balance < 0 is precisely how we decide which card holds to settle.

[chadwhitacre]

One possibility would be to floor old_balance and/or new_balance at 0. Would that work cleanly?

[chadwhitacre]

Alright, this is non-trivial.

[chadwhitacre]

Spurious build failures? Restarted ...

[chadwhitacre]

Rebased, was 68e36a2.

[chadwhitacre]

Ping @mattbk @nobodxbodon @kaguillera I think this should be an easy one to review, because it only touches tests.

[chadwhitacre]

Withdrawing from review. Tests are failing and it's not obvious to me why.

[chadwhitacre]

Giving up. 😞