Near-term Game Plan

• Get basic mid-level tests working. Don't worry about making them comprehensive -- just get to the point where we can implement regression tests this way quickly and easily.
• Get export/import of Spaces working again. Don't get super-fancy: just export the current snapshot format, in serialized binary, and import that.
• Improve the TagSet UI. This should be low-hanging fruit: add an API call to fetch all of the values for the given Tags, and then build a drop-down to the left of the Manifest UI, that lets you pick items from there and add them to the Manifest automatically.
• Get Experiment Mode starting to work. In theory, this builds on the existing Publication infrastructure.
• Build the proper Choice UI. This is a dramatic rewrite for editing an old-fashioned Choice value. This should be entirely client-rendered, as opposed to how it currently works. The actual display depends on the number of items -- radio buttons for a short list, drop-down plus prompting text for a longer one. In all cases, the HTML just specifies the top level: even in the radio-button case, we query the API to fetch the values. You can specify in _edit() if you prefer a specific option; that overrides the default. But the current drop-down-only should just plain go away.
• Figure out the game plan for replacing ConductR: it's increasingly proving a deadly bottleneck. I suspect the right answer for now is to pick up Mesos or Kubernetes, but I'm not at all sure about that.