fix(stats): include execution errors in pass/total denominator (#999)

Previously the summary showed passedCount/gradedCount where gradedCount
excluded execution errors, making results appear better than they were.
Now shows passedCount/total to match the Convex Evals convention that
inspired this design: the denominator is all tests actually attempted,
not just those that reached the grading stage.

Execution errors are still reported separately in the detail section.

Closes #998

Author: christso

apps/cli/src/commands/eval/statistics.ts

@@ -226,7 +226,7 @@ export function formatEvaluationSummary(
   } else {
     overallVerdict = overallPassed ? 'PASS' : 'FAIL';
     verdictColor = overallPassed ? '\x1b[32m' : '\x1b[31m';
-    verdictText = `RESULT: ${overallVerdict}  (${summary.passedCount}/${gradedCount} scored >= ${threshold}, mean: ${formatScore(summary.mean)})`;
+    verdictText = `RESULT: ${overallVerdict}  (${summary.passedCount}/${summary.total} scored >= ${threshold}, mean: ${formatScore(summary.mean)})`;
   }
 
   lines.push('\n==================================================');