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>

config.yaml

@@ -18,6 +18,7 @@ sentiment:
   rolling_window_size: 10  # Number of messages to track per user
   rolling_window_minutes: 15  # Time window for tracking
   batch_window_seconds: 3  # Wait this long for more messages before analyzing (debounce)
+  escalation_threshold: 0.25  # Triage toxicity score that triggers re-analysis with heavy model
 
 game_channels:
   gta-online: "GTA Online"