fix(ai): normalize boolean scores in onlineEval scoresSummary (#263)

## Overview

- `onlineEval()` was writing raw boolean scores (`true`/`false`) into
the parent eval span's `eval.case.scores` attribute, while child scorer
spans correctly normalized them to `1`/`0` with `eval.score.is_boolean`
metadata via `normalizeBooleanScore()`
- Apply the same `normalizeBooleanScore()` call when building
`scoresSummary` so both parent and child spans produce consistent
numeric scores

<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Low Risk**
> Small telemetry-only change that affects how scores are serialized
into span attributes; low risk aside from potential downstream
expectations of boolean values.
> 
> **Overview**
> Ensures `onlineEval()` writes consistent numeric scores into the
parent eval span’s `eval.case.scores` summary by normalizing boolean
`score` values (`true/false` → `1/0`) and propagating the corresponding
`eval.score.is_boolean` metadata.
> 
> This updates `onlineEval.ts` to call `normalizeBooleanScore()` while
building `scoresSummary`, and only emits normalized metadata when
non-empty.
> 
> <sup>Written by [Cursor
Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit
bfa6ce7. This will update automatically
on new commits. Configure
[here](https://cursor.com/dashboard?tab=bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

Author: lukasmalkmus

packages/ai/src/online-evals/onlineEval.ts

@@ -9,6 +9,7 @@ import type {
   ScorerSampling,
 } from './types';
 import { executeScorer } from './executor';
+import { normalizeBooleanScore } from '../evals/normalize-score';
 import { Attr } from '../otel/semconv/attributes';
 import type { ValidateName } from '../util/name-validation';
 import { isValidName } from '../util/name-validation-runtime';
@@ -359,11 +360,16 @@ async function executeOnlineEvalInternal<
 
     const scoresSummary: Record<string, ScorerResult> = {};
     for (const [name, result] of Object.entries(results)) {
+      const { score: normalizedScore, metadata: normalizedMetadata } = normalizeBooleanScore(
+        result.score,
+        result.metadata,
+      );
+
       scoresSummary[name] = {
         name: result.name,
-        score: result.score,
-        ...(result.metadata &&
-          Object.keys(result.metadata).length > 0 && { metadata: result.metadata }),
+        score: normalizedScore,
+        ...(normalizedMetadata &&
+          Object.keys(normalizedMetadata).length > 0 && { metadata: normalizedMetadata }),
         ...(result.error && { error: result.error }),
       };
     }