perf(brain): switch voice brain to Sonnet 5 with an always-on brevity rule

Sonnet 5 reaches first token ~0.4s sooner than Sonnet 4.5 but answers the
same voice prompt more verbosely (measured 46-54 vs ~28 output tokens),
which erased the win in total turn time. Add a cached, persona-independent
brevity system block that pulls output back to ~30 tokens, so the faster
first token becomes a faster, lower-variance whole reply.

Measured (16-round interleaved A/B, production-shaped call):
  sonnet-4-5      TTFT 1.26s  total 2.02s (tail 3.26s)  out 29
  sonnet-5+brev   TTFT 0.85s  total 1.70s (tail 2.28s)  out 30

- ClaudeBrain: default model claude-sonnet-5 + BREVITY block (cached with
  the persona prefix so a dashboard persona edit can't drop it).
- Defaults aligned: config.Settings.anthropic_model and the voice-server
  WSAI_BRAIN_MODEL default -> claude-sonnet-5.
- Dashboard: add claude-sonnet-5 to LLM_OPTIONS + JS label; fix stale
  restart hint.
- tests/latency_ab.py: reproducible model-latency A/B harness (reads
  CLAUDE_CREDENTIALS_PATH; makes live API calls, so not a pytest test).

Streaming TTS was intentionally not added: this bot answers in one sentence,
where sentence-level streaming has no overlap to exploit, and it would
require rearchitecting both the Python endpoint and the node playback.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

wsai/backends/claude.py

@@ -127,6 +127,15 @@ class ClaudeVision:
 
 
 class ClaudeBrain:
+    # Injected as an always-on, cached system block on top of the (editable)
+    # persona. Sonnet 5 answers correctly but more verbosely than 4.5 for the
+    # same voice prompt (measured 46-54 vs 28 output tokens), which erased its
+    # ~0.4s time-to-first-token advantage in total turn time. This hard brevity
+    # rule pulls Sonnet 5 back to ~30 tokens, so the faster first token actually
+    # translates into a faster (and lower-variance) whole reply. Kept separate
+    # from PERSONA so a dashboard persona edit can never drop it.
+    BREVITY = "지금부터 답은 무조건 한 문장, 12단어 이내로만. 부연·재확인·군더더기 금지."
+
     PERSONA = (
         "너는 디스코드를 이용해 사용자와 실시간으로 대화하는 AI 인공지능이야.\n\n"
         "1. 역할\n"
@@ -163,7 +172,7 @@ class ClaudeBrain:
         "- 너는 디스코드에서 함께 대화하는 실시간 AI 인공지능이다."
     )
 
-    def __init__(self, *, model: str = "claude-sonnet-4-5", api_key: str | None = None) -> None:
+    def __init__(self, *, model: str = "claude-sonnet-5", api_key: str | None = None) -> None:
         self.model = model
         self._auth = _Auth(api_key)
 
@@ -176,8 +185,9 @@ class ClaudeBrain:
         msgs.append({"role": "user", "content": screen_note + user_text})
         client = self._auth.client()
         # Read the persona live each turn so a dashboard edit applies immediately
-        # (falls back to the built-in PERSONA when no override is saved).
-        system = self._auth.system(get_persona(self.PERSONA))
+        # (falls back to the built-in PERSONA when no override is saved). The
+        # brevity rule is appended as its own block so it survives persona edits.
+        system = self._auth.system(get_persona(self.PERSONA), self.BREVITY)
         if system:
             # Cache the (static) system prompt so repeat turns skip re-processing
             # it — lower time-to-first-token. No-op below the model's cache