feat: 重做阅读找答案题与 AI 情境对话，并归拢本地识别、查词等既有改动

阅读"在对话里找到答案"：
- 16 道题全部重写，干扰项真实出现在对话里，靠说话人归属或否定句才能作答
- 选项按题目内容确定性打乱，答案不再固定排第一；听力环节同样处理
- 去掉超纲干扰项、重复题干，收紧自由作答匹配（过去单个字母也能判对）

AI 情境对话：
- 提示词区分"AI 这一句要做什么"与"学习者随后要完成什么"，并下发已教词句清单
- JSON 只强制 reply，translation/feedback 可选；不再索要用不上的 slots/evidence
- AI 不可用时页面明确提示当前回复来自内置示范脚本
- 删掉按 stage 下标猜中文翻译的兜底，避免译文与英文对不上
- 整课对话改用逐轮必需表达校验，替换"关键词沾边就算过"；修正自由场景正则误伤
- 总结的"完成任务"按实际通过的轮次生成；模型点评只在结束页呈现一次
- 自由场景支持草稿续练（独立存储槽）；修正回答轮数文案与永不解锁的场景标注

同时提交此前工作区中累积的改动：SenseVoice 本地识别、查词/句型解析卡、
复习与测评页调整等，并补充对话校验、选项分布和句子解析的测试。

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

kouyu_english/test/sherpa_stt_test.dart

@@ -1,20 +1,32 @@
+// ignore_for_file: avoid_print
 import 'dart:io';
 import 'package:flutter_test/flutter_test.dart';
+import 'package:kouyu_english/core/sherpa_stt_service.dart';
 import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
 
 void main() {
   TestWidgetsFlutterBinding.ensureInitialized();
 
+  final home = Platform.environment['HOME'] ?? '';
+  final pubCache = Platform.environment['PUB_CACHE'] ?? '$home/.pub-cache';
+  final candidateDirs = [
+    '$pubCache/hosted/pub.dev/sherpa_onnx_macos-1.13.8/macos',
+    '$home/.pub-cache/hosted/pub.dev/sherpa_onnx_macos-1.13.8/macos',
+    '/Users/shenlei/.pub-cache/hosted/pub.dev/sherpa_onnx_macos-1.13.8/macos',
+  ];
+  final macosDir = candidateDirs.firstWhere(
+    (d) => Directory(d).existsSync(),
+    orElse: () => candidateDirs.first,
+  );
+
   test('SenseVoice-Small ONNX transcribes WAV audio file accurately', () async {
-    const macosDir = '/Users/shen/.pub-cache/hosted/pub.dev/sherpa_onnx_macos-1.13.8/macos';
-    
     // Test direct Sherpa ASR initialization and decoding with SenseVoice
     sherpa_onnx.initBindings(macosDir);
 
     final modelConfig = sherpa_onnx.OfflineModelConfig(
       senseVoice: const sherpa_onnx.OfflineSenseVoiceModelConfig(
         model: 'assets/models/sense_voice/model.int8.onnx',
-        language: 'auto',
+        language: 'en',
         useInverseTextNormalization: true,
       ),
       tokens: 'assets/models/sense_voice/tokens.txt',
@@ -38,10 +50,26 @@ void main() {
       stream.acceptWaveform(samples: wave.samples, sampleRate: wave.sampleRate);
       recognizer.decode(stream);
       final result = recognizer.getResult(stream);
-      print("SenseVoice transcribed result: ${result.text}");
-      expect(result.text.toLowerCase().contains("nightfall"), isTrue);
+      print('SenseVoice transcribed result: ${result.text}');
+      expect(result.text.toLowerCase().contains('nightfall'), isTrue);
       stream.free();
     }
     recognizer.free();
   });
+
+  test('SherpaSttService singleton initializes and transcribes cleanly', () async {
+    final ready = await SherpaSttService.instance.initialize(nativeLibDir: macosDir);
+    expect(ready, isTrue);
+    expect(SherpaSttService.instance.isReady, isTrue);
+
+    const testWave = '/tmp/sherpa_test/sherpa-onnx-zipformer-small-en-2023-06-26/test_wavs/0.wav';
+    if (File(testWave).existsSync()) {
+      final transcribed = await SherpaSttService.instance.transcribeWav(testWave);
+      expect(transcribed, isNotNull);
+      print('SherpaSttService transcribed: $transcribed');
+      expect(transcribed!.toLowerCase().contains('nightfall'), isTrue);
+      // SenseVoice tags should be stripped
+      expect(transcribed.contains('<|'), isFalse);
+    }
+  });
 }