Implement most-general BRAVO procedure

[ranweiler]

The algorithm we have right now is not general, and is only appropriate for a plurality contest with at most two candidates, or majority contests with any number of candidates. Replace this initial implementation with the general C, k procedure in the BRAVO paper (section 7). This is appropriate for plurality contests with many candidates and even many winners.

It also makes sense, as part of this, to track all risk and ASN computations in the backend. This way they never need to be recomputed when fetched, and we can have a cache/trace of the progress of the audit. Ensure that we do this in a way that supports auditors updating previously-marked ballots, so we can have a full log of interesting audit events.