feat: 完善听力口语辅助状态记录、练习分块与复习抽查间隔机制

kouyu_english/lib/core/models.dart

@@ -92,6 +92,8 @@ enum EvidenceKind {
   pending,
 }
 
+enum ListeningSupport { none, slow, transcript }
+
 /// Local dialogue matching is intentionally tri-state. A natural English
 /// reply that does not match a strict taught-language rule is unknown, not
 /// automatically wrong.
@@ -160,6 +162,10 @@ class AttemptEvidence {
     this.originalTranscript,
     this.transcriptConfirmed = false,
     this.transcriptEdited = false,
+    this.listeningSupport,
+    this.firstTry,
+    this.attemptCount,
+    this.usedSentenceBuild = false,
   });
 
   final String id;
@@ -182,6 +188,10 @@ class AttemptEvidence {
   final String? originalTranscript;
   final bool transcriptConfirmed;
   final bool transcriptEdited;
+  final ListeningSupport? listeningSupport;
+  final bool? firstTry;
+  final int? attemptCount;
+  final bool usedSentenceBuild;
 }
 
 enum AiProviderType { mock, gemini, openAi, compatible }
@@ -212,8 +222,9 @@ class DialogueAiResponse {
   final String? feedback;
 }
 
-/// A bounded suggestion only. It never changes completion, mastery, or a
-/// stage result; those decisions remain on-device and evidence-based.
+/// A bounded evaluation input. Lesson free-response evidence requires both
+/// local task matching and an accepted verdict; mastery and stage decisions
+/// still remain on-device and derive from recorded attempts.
 class WritingAiFeedback {
   const WritingAiFeedback({
     required this.verdict,