(autofix) Improve root cause scoring prompt

[aayush-se]

[jennmueng]

Not really any better way than collecting more examples for the eval set, we can add LLM-as-a-judge metrics using Braintrust. We can also consider a thumbs up/down on the root cause too.