feat!: Rename JudgeResponse to JudgeResult and flatten EvalScore

[jsonbailey]

Summary

• Replaces JudgeResponse + nested EvalScore dict with a flat JudgeResult dataclass — one judge produces one result, so the dict was unnecessary abstraction
• Adds sampled: bool to cleanly distinguish skipped-by-sampling-rate from failure (previously returned None)
• Judge.evaluate() always returns a JudgeResult — never None; judge_config_key is always set on every return path
• Renames track_judge_response → track_judge_result on LDAIConfigTracker; removes track_eval_scores
• Removes track_judge_response from AIGraphTracker — judges are node-level only (spec updated in launchdarkly/sdk-specs#147)

Breaking Changes

• JudgeResponse removed — use JudgeResult
• EvalScore removed — fields are now inline on JudgeResult (score, reasoning, metric_key)
• error field renamed to error_message
• track_judge_response removed — use track_judge_result
• track_eval_scores removed
• Judge.evaluate() returns JudgeResult instead of Optional[JudgeResponse]

Test plan

• All 114 server-ai tests passing

🤖 Generated with Claude Code

---

Note

Medium Risk
Breaking API change to judge evaluation return types and tracking hooks; may impact downstream consumers expecting None/dict-shaped evaluations or calling removed tracker methods.

Overview
Simplifies judge evaluation results and tracking. Replaces JudgeResponse + nested EvalScore dict with a single flat JudgeResult (score/reasoning/metric key), and updates exports so JudgeResult is the public type.

Judge.evaluate()/evaluate_messages() now always return a JudgeResult (never None), using sampled=True to represent sampling skips and error_message for failures. ModelResponse.evaluations and ManagedModel’s judge dispatch/tracking are updated accordingly, and LDAIConfigTracker consolidates judge metric emission into track_judge_result while removing the old track_eval_scores/track_judge_response paths; tests are updated to match the new contract.

^{Reviewed by Cursor Bugbot for commit 7e23fa2. Bugbot is set up for automated code reviews on this repo. Configure here.}