fix: inject argument topic into EVERY exchange, not just the first message

The topic was only being injected into the initial breakthrough message via
get_argument_start_prompt(). After that, every subsequent exchange called
get_miku_argument_prompt() / get_evil_argument_prompt() which had no concept
of the topic — so both personas forgot what they were arguing about after the
first exchange and reverted to generic identity-crisis arguments.

Fix: added argument_topic parameter to both persona prompt functions and inject
it as a bold ARGUMENT THEME reminder in every single exchange. The topic block
explicitly tells the LLM to stay on-topic and not drift into generic territory.

bot/utils/bipolar_mode.py

@@ -764,7 +764,7 @@ def _get_mood_argument_guidance(persona: str) -> str:
     return ""
 
 
-def get_miku_argument_prompt(evil_message: str, context: str = "", is_first_response: bool = False, argument_history: str = "") -> str:
+def get_miku_argument_prompt(evil_message: str, context: str = "", is_first_response: bool = False, argument_history: str = "", argument_topic: str = "") -> str:
     """Get prompt for Regular Miku to respond in an argument"""
     if is_first_response:
         message_context = f"""You just noticed something Evil Miku said in the chat:
@@ -788,9 +788,19 @@ ARGUMENT SO FAR (DO NOT REPEAT THESE POINTS):
 You already made your points above. Now respond to her LATEST message specifically. 
 Do NOT rehash what you've already said — push the argument FORWARD with new angles."""
     
+    # Build topic reminder — keeps the argument on-theme
+    topic_block = ""
+    if argument_topic:
+        topic_block = f"""
+
+ARGUMENT THEME: {argument_topic}
+This is what you're arguing about. Stay on THIS topic. Every response should connect back to this theme.
+Do NOT drift into generic "who's the real Miku" territory — stick to THIS specific subject."""
+    
     return f"""You are Hatsune Miku responding in an argument with your evil alter ego.
 {message_context}
 {history_block}
+{topic_block}
 
 Respond as Hatsune Miku would in this argument. You're NOT just meek and frightened - you're the REAL Miku, 
 and you have every right to stand up for yourself and defend who you are. While you're generally kind and 
@@ -808,7 +818,7 @@ Don't use any labels or prefixes.
 Your current mood is: {globals.DM_MOOD}"""
 
 
-def get_evil_argument_prompt(miku_message: str, context: str = "", is_first_response: bool = False, argument_history: str = "") -> str:
+def get_evil_argument_prompt(miku_message: str, context: str = "", is_first_response: bool = False, argument_history: str = "", argument_topic: str = "") -> str:
     """Get prompt for Evil Miku to respond in an argument"""
     if is_first_response:
         message_context = f"""You just noticed something Regular Miku said in the chat:
@@ -832,9 +842,19 @@ ARGUMENT SO FAR (DO NOT REPEAT THESE POINTS):
 You already made your points above. Now respond to her LATEST message specifically. 
 Do NOT rehash what you've already said — push the argument FORWARD with new, sharper angles."""
     
+    # Build topic reminder — keeps the argument on-theme
+    topic_block = ""
+    if argument_topic:
+        topic_block = f"""
+
+ARGUMENT THEME: {argument_topic}
+This is what you're arguing about. Stay on THIS topic. Every response should connect back to this theme.
+Do NOT drift into generic "who's the real Miku" territory — stick to THIS specific subject."""
+    
     return f"""You are Evil Miku responding in an argument with your "good" counterpart.
 {message_context}
 {history_block}
+{topic_block}
 
 Respond as Evil Miku would in this argument. You're not just mindlessly cruel - you're CALCULATING, 
 intelligent, and strategic. You know how to get under her skin and you're the DARK reflection of everything 
@@ -1443,9 +1463,9 @@ Your current mood is: {globals.EVIL_DM_MOOD if loser == 'evil' else globals.DM_M
             
             # Generate response with context about what the other said
             if current_speaker == "evil":
-                response_prompt = get_evil_argument_prompt(last_message, is_first_response=is_first_response, argument_history=arg_history)
+                response_prompt = get_evil_argument_prompt(last_message, is_first_response=is_first_response, argument_history=arg_history, argument_topic=argument_topic)
             else:
-                response_prompt = get_miku_argument_prompt(last_message, is_first_response=is_first_response, argument_history=arg_history)
+                response_prompt = get_miku_argument_prompt(last_message, is_first_response=is_first_response, argument_history=arg_history, argument_topic=argument_topic)
             
             # Use force_evil_context to avoid race condition with globals.EVIL_MODE
             response = await query_llama(