feat: add pick_reaction method to LLMClient

Lightweight LLM call that picks a contextual emoji reaction for a
Discord message. Uses temperature 0.9 for variety, max 16 tokens,
and validates the response is a short emoji token or returns None.

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

utils/llm_client.py

@@ -743,6 +743,61 @@ class LLMClient:
                 self._log_llm("classify_intent", elapsed, False, message_text[:200], error=str(e))
                 return "chat"
 
+    async def pick_reaction(self, message_text: str, channel_name: str) -> str | None:
+        """Pick a contextual emoji reaction for a Discord message.
+
+        Returns an emoji string, or None if no reaction is appropriate.
+        """
+        prompt = (
+            "You are a lurker in a Discord gaming server. "
+            "Given a message and its channel, decide if it deserves a reaction emoji.\n\n"
+            "Available reactions:\n"
+            "\U0001f480 = funny/dead\n"
+            "\U0001f602 = hilarious\n"
+            "\U0001f440 = drama/spicy\n"
+            "\U0001f525 = impressive\n"
+            "\U0001f4af = good take\n"
+            "\U0001f62d = sad/tragic\n"
+            "\U0001f921 = clown moment\n"
+            "\u2764\ufe0f = wholesome\n"
+            "\U0001fae1 = respect\n"
+            "\U0001f913 = nerd\n"
+            "\U0001f974 = drunk/unhinged\n"
+            "\U0001f3af = accurate\n\n"
+            "Reply with ONLY the emoji, or NONE if the message doesn't warrant a reaction. "
+            "Most messages should get NONE — only react when something genuinely stands out."
+        )
+        t0 = time.monotonic()
+
+        async with self._semaphore:
+            try:
+                temp_kwargs = {"temperature": 0.9} if self._supports_temperature else {}
+                response = await self._client.chat.completions.create(
+                    model=self.model,
+                    messages=[
+                        {"role": "system", "content": prompt},
+                        {"role": "user", "content": f"[#{channel_name}] {message_text}"},
+                    ],
+                    **temp_kwargs,
+                    max_completion_tokens=16,
+                )
+                elapsed = int((time.monotonic() - t0) * 1000)
+                raw = (response.choices[0].message.content or "").strip()
+                token = raw.split()[0] if raw.split() else ""
+
+                if not token or token.lower() == "none" or len(token) > 7:
+                    self._log_llm("pick_reaction", elapsed, True, message_text[:200], "NONE")
+                    return None
+
+                self._log_llm("pick_reaction", elapsed, True, message_text[:200], token)
+                logger.debug("Picked reaction %s for: %s", token, message_text[:80])
+                return token
+            except Exception as e:
+                elapsed = int((time.monotonic() - t0) * 1000)
+                logger.error("Reaction pick error: %s", e)
+                self._log_llm("pick_reaction", elapsed, False, message_text[:200], error=str(e))
+                return None
+
     async def extract_memories(
         self,
         conversation: list[dict[str, str]],