/roborev:fix gets confused about analyze task results

[wesm]

In Claude Code, it searches for a verdict and can't find one, and then disregards the findings. This differs from roborev fix on the CLI which has no problem processing refactoring from the analyze requests

[mariusvniekerk]

We probably want to have different categories of these roborev initiated runs (as part of the data model), knowing explicitly if something is a review vs analyze will help things like fix know if they should treat it as a review or not.