feat: full feature buildout — streaming, i18n, mastery map, admin, jobs

Progressive lesson streaming via onSegment callback (fixes SSE for non-English
users — locale was shadowed in lesson-reader useEffect). Adds: BullMQ workers,
Redis stream buffer, token budget enforcement, Langfuse tracing, golden-eval
runner, Playwright e2e scaffolding, lesson depth/locale/preferences schema,
mastery map UI, admin panel (blueprints/users/reports/quality/misconceptions),
image queries, source citations, view transitions, reading animations, i18n
(next-intl), PDF export, surprise endpoint, and 402 passing unit tests.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

tests/golden/run-grading-eval.ts

@@ -187,6 +187,25 @@ async function main(): Promise<void> {
     console.log('\n✓ PASS: grading quality within thresholds\n');
   }
 
+  // Persist run to DB when PERSIST_EVAL_RUNS=1 (CI/admin use; skip locally by default)
+  if (process.env.PERSIST_EVAL_RUNS === '1') {
+    try {
+      const { insertEvalRun } = await import('../../src/lib/db/queries');
+      await insertEvalRun({
+        suite: 'grading',
+        falseFailRate,
+        accuracy,
+        total: CASES.length,
+        passed: Math.round(accuracy * CASES.length),
+        modelVersion: process.env.LLM_GRADER_MODEL ?? 'unknown',
+        gitSha: process.env.GIT_SHA,
+      });
+      console.log('[eval] Run persisted to eval_run table.');
+    } catch (err) {
+      console.warn('[eval] Failed to persist run (non-fatal):', err);
+    }
+  }
+
   process.exit(failed ? 1 : 0);
 }