feat: fix evil mode race conditions, expand moods and PFP detection

bipolar_mode.py:
- Replace unsafe globals.EVIL_MODE temporary overrides with
  force_evil_context parameter to fix async race conditions (3 sites)

moods.py:
- Add 6 new evil mood emojis: bored, manic, jealous, melancholic,
  playful_cruel, contemptuous
- Refactor rotate_dm_mood() to skip when evil mode active (evil mode
  has its own independent 2-hour rotation timer)

persona_dialogue.py:
- Same force_evil_context race condition fix (2 sites)
- Fix over-aggressive response cleanup that stripped common words
  (YES/NO/HIGH) — now uses targeted regex for structural markers only
- Update evil mood multipliers to match new mood set

profile_picture_context:
- Expand PFP detection regex for broader coverage (appearance questions,
  opinion queries, selection/change questions)
- Add plugin.json metadata file

bot/utils/bipolar_mode.py

@@ -883,23 +883,15 @@ async def run_argument(channel: discord.TextChannel, client, trigger_context: st
         if last_message is None:
             init_prompt = get_argument_start_prompt(initiator, trigger_context)
             
-            # Temporarily set evil mode for query_llama if initiator is evil
-            original_evil_mode = globals.EVIL_MODE
-            if initiator == "evil":
-                globals.EVIL_MODE = True
-            else:
-                globals.EVIL_MODE = False
-            
-            try:
-                initial_message = await query_llama(
-                    user_prompt=init_prompt,
-                    user_id=argument_user_id,
-                    guild_id=guild_id,
-                    response_type="autonomous_general",
-                    model=globals.EVIL_TEXT_MODEL if initiator == "evil" else globals.TEXT_MODEL
-                )
-            finally:
-                globals.EVIL_MODE = original_evil_mode
+            # Use force_evil_context to avoid race condition with globals.EVIL_MODE
+            initial_message = await query_llama(
+                user_prompt=init_prompt,
+                user_id=argument_user_id,
+                guild_id=guild_id,
+                response_type="autonomous_general",
+                model=globals.EVIL_TEXT_MODEL if initiator == "evil" else globals.TEXT_MODEL,
+                force_evil_context=(initiator == "evil")
+            )
             
             if not initial_message or initial_message.startswith("Error") or initial_message.startswith("Sorry"):
                 logger.error("Failed to generate initial argument message")
@@ -994,23 +986,15 @@ async def run_argument(channel: discord.TextChannel, client, trigger_context: st
                     # Add last message as context
                     response_prompt = f'The other Miku said: "{last_message}"\n\n{end_prompt}'
                     
-                    # Temporarily set evil mode for query_llama
-                    original_evil_mode = globals.EVIL_MODE
-                    if winner == "evil":
-                        globals.EVIL_MODE = True
-                    else:
-                        globals.EVIL_MODE = False
-                    
-                    try:
-                        final_message = await query_llama(
-                            user_prompt=response_prompt,
-                            user_id=argument_user_id,
-                            guild_id=guild_id,
-                            response_type="autonomous_general",
-                            model=globals.EVIL_TEXT_MODEL if winner == "evil" else globals.TEXT_MODEL
-                        )
-                    finally:
-                        globals.EVIL_MODE = original_evil_mode
+                    # Use force_evil_context to avoid race condition with globals.EVIL_MODE
+                    final_message = await query_llama(
+                        user_prompt=response_prompt,
+                        user_id=argument_user_id,
+                        guild_id=guild_id,
+                        response_type="autonomous_general",
+                        model=globals.EVIL_TEXT_MODEL if winner == "evil" else globals.TEXT_MODEL,
+                        force_evil_context=(winner == "evil")
+                    )
                     
                     if final_message and not final_message.startswith("Error") and not final_message.startswith("Sorry"):
                         # Send winner's final message via webhook
@@ -1059,23 +1043,15 @@ async def run_argument(channel: discord.TextChannel, client, trigger_context: st
             else:
                 response_prompt = get_miku_argument_prompt(last_message, is_first_response=is_first_response)
             
-            # Temporarily set evil mode for query_llama
-            original_evil_mode = globals.EVIL_MODE
-            if current_speaker == "evil":
-                globals.EVIL_MODE = True
-            else:
-                globals.EVIL_MODE = False
-            
-            try:
-                response = await query_llama(
-                    user_prompt=response_prompt,
-                    user_id=argument_user_id,
-                    guild_id=guild_id,
-                    response_type="autonomous_general",
-                    model=globals.EVIL_TEXT_MODEL if current_speaker == "evil" else globals.TEXT_MODEL
-                )
-            finally:
-                globals.EVIL_MODE = original_evil_mode
+            # Use force_evil_context to avoid race condition with globals.EVIL_MODE
+            response = await query_llama(
+                user_prompt=response_prompt,
+                user_id=argument_user_id,
+                guild_id=guild_id,
+                response_type="autonomous_general",
+                model=globals.EVIL_TEXT_MODEL if current_speaker == "evil" else globals.TEXT_MODEL,
+                force_evil_context=(current_speaker == "evil")
+            )
             
             if not response or response.startswith("Error") or response.startswith("Sorry"):
                 logger.error(f"Failed to generate argument response")