Currently we bisect into the greater of two regressions at any step.

We should do something to handle cases were more than one of them appear on the same range:

A doc should be written do discuss what the best approach is, but from a high-level perspective some options are:
- Output possible regression ranges and their approximate magnitudes in the partial and final results, and let the dashboard decide what to do (if anything).
- Decide on the bisect itself if a separate bisect job should be spawned, and trigger it.
- Make bisect jobs handle multiple culrpits, and make a single job find all of them.