fix: 修正学习流程中的证据、复习与进度问题

- 独立尝试按课程要求校验内容，不再任意输入即算独立成功
- 复习节点改为学后第 1/3/7/14 天，之后每 30 天抽查；复习轮换情境
- 掌握状态按作答记录推导：需认识、可回忆、跨情境使用及最近两次无帮助
- 课程证据按实际用到的目标记录；跟读与课程对话只算辅助练习
- 复习成功不再降低已有状态
- 学完课程后进入下一节未完成课程；全部学完后引导巩固与阶段评估
- 新增 3 题基础定位，按结果设置起始课程
- 新增按课程开放的对话场景，首页推荐最少练习的场景
- 切换课程时清空上一课的步骤进度
- AI 返回 JSON 格式并按 JSON 回放历史；拼写校验忽略大小写与连字符；移除学习目标选择

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

kouyu_english/lib/core/ai_service.dart

@@ -382,12 +382,19 @@ class AiService {
     // with a small token cap, so reasoning only adds cost and can exhaust the
     // cap before the JSON is written. Other providers may reject these
     // DeepSeek-specific switches, so they keep the original payload.
+    //
+    // Every prompt asks for JSON, but without JSON Output DeepSeek copies the
+    // plain-text assistant turns in a dialogue history and answers in plain
+    // text about half the time, which the app then treats as a failed call.
     if (_isDeepSeek(uri)) {
       return isResponses
           ? {
               'model': model,
               'input': messages,
               'reasoning': {'effort': 'none'},
+              'text': {
+                'format': {'type': 'json_object'},
+              },
               'temperature': ?temperature,
               'max_output_tokens': ?maxTokens,
             }
@@ -395,6 +402,7 @@ class AiService {
               'model': model,
               'messages': messages,
               'thinking': {'type': 'disabled'},
+              'response_format': {'type': 'json_object'},
               'temperature': ?temperature,
               'max_tokens': ?maxTokens,
             };
@@ -787,7 +795,18 @@ class AiService {
       endpoint: endpoint,
       model: model,
       system: system,
-      messages: history,
+      // Earlier AI lines are stored as plain English. Sent that way they teach
+      // the model to answer in plain text (or, in JSON mode, with blanks), so
+      // they are replayed in the JSON shape the system prompt asks for.
+      messages: [
+        for (final message in history)
+          message['role'] == 'assistant'
+              ? {
+                  'role': 'assistant',
+                  'content': jsonEncode({'reply': message['content']}),
+                }
+              : message,
+      ],
       temperature: 0.3,
       maxTokens: 300,
     );