Automated issue advancement

At a high level:
* Undesigned issues are automatically picked up by an iterative design workflow
* Designed *and* planned issues are automatically picked up by an implementation workflow

Those workflows must:
* Have a hard iteration ceiling to stop runaway bad plans
* Be focussed on minimising human review time. People should be able to look at a thoroughly reviewed and AI approved PR with a high level of confidence that it will be mergeable the first time around. This won't happen overnight, but it's the north star in any design decision/prompt evaluation.
* Be token/model/cost efficient.
* Have a manageable DOP. Maybe this means pausing when there's X human-ready PR. Maybe human review throughput is it's own backpressure (though that sounds mentally overwhelming).
* Have appropriate stop conditions. For implementation this is easier, since we have a thorough change proposal. Thought needs to be put into when a design proposal is human ready.

The issue names no files, tests, or entry points. Start by documenting the design and implementation workflows described in the issue, then define their iteration ceilings, stop conditions, review backpressure, and cost constraints. Done means the workflows have an agreed, reviewable design.

Written by the indexing model from the issue text.