Optimisation performance (30t/s --> 50t/s) & adaptation parsing json --> llama.cpp & Fix système mémoire

ia/memoire.py

@@ -408,6 +408,41 @@ async def build_context_for_model(
     return joined
 
 
+async def get_structured_history(
+    user_id: int | str,
+    max_recent: int = 12,
+    guild_id: str | None = None,
+) -> tuple[str, List[Dict[str, str]]]:
+    """Retourne le résumé global et l'historique structuré (utile pour l'API Chat)."""
+    data = await load_user_memory_async(user_id, guild_id)
+    resume = data.get("resume_global", "")
+    
+    recent = data.get("historique_recent", [])
+    recent_tail = recent[-max_recent:] if recent else []
+    
+    structured = []
+    import datetime
+    for m in recent_tail:
+        um = m.get("user_message", "").strip()
+        bm = m.get("bot_message", "").strip()
+        ts = m.get("timestamp")
+        
+        time_prefix = ""
+        if ts:
+            try:
+                time_str = datetime.datetime.fromtimestamp(float(ts)).strftime("%H:%M:%S")
+                time_prefix = f"[{time_str}] "
+            except (ValueError, TypeError):
+                pass
+            
+        if um:
+            structured.append({"role": "user", "content": f"{time_prefix}{um}"})
+        if bm:
+            structured.append({"role": "assistant", "content": f"{time_prefix}{bm}"})
+            
+    return resume, structured
+
+
 async def purge_old_memories(days: int = 30) -> int:
     """
     Supprime les fichiers de mémoire plus vieux que X jours pour conformité RGPD.