Add two-tier LLM analysis with triage/escalation

Triage model (LLM_MODEL) handles every message cheaply. If toxicity
>= 0.25, off_topic, or coherence < 0.6, the message is re-analyzed
with the heavy model (LLM_ESCALATION_MODEL). Chat, image analysis,
/bcs-test, and /bcs-scan always use the heavy model.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

cogs/chat.py

@@ -84,7 +84,7 @@ class ChatCog(commands.Cog):
                 image_attachment.filename,
                 user_text[:80],
             )
-            response = await self.bot.llm.analyze_image(
+            response = await self.bot.llm_heavy.analyze_image(
                 image_bytes,
                 SCOREBOARD_ROAST,
                 user_text=user_text,
@@ -108,7 +108,7 @@ class ChatCog(commands.Cog):
                 {"role": "user", "content": f"{score_context}\n{message.author.display_name}: {content}"}
             )
 
-            response = await self.bot.llm.chat(
+            response = await self.bot.llm_heavy.chat(
                 list(self._chat_history[ch_id]),
                 CHAT_PERSONALITY,
                 on_first_token=start_typing,