fix(tts_say): read tts_model from config (was kokoro_model) + WAV format

Bug: skill was reading config key "kokoro_model" but ~/.codec/config.json
stores it under "tts_model". Fell back to literal string "kokoro" which
mlx-audio rejects with HTTP 500, silently degrading to macOS `say` voice
on every MCP tts_say invocation.

Fix:
- Read tts_model first, fall back to legacy kokoro_model, then sane default
- Request response_format: wav explicitly
- Save tempfile with .wav extension (mlx-audio returns PCM WAV at 24kHz)
- Bump timeout 20s -> 30s for first-token latency on cold model

Verified: Kokoro TTS now plays real Kokoro voice (bm_george/am_adam/af_bella)
on both direct CLI test and MCP invocation from claude.ai.

Also fixed during same sweep (n8n, not in repo):
- Daily Report.MF workflow had YOUR_LOCAL_IP placeholder -> 192.168.1.73
- Lucy QWEN 5.1/5.2 workflows pointed doc-extract at kokoro port -> 8086

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

Author: Mikarina13

skills/tts_say.py

@@ -17,7 +17,8 @@
     _cfg = {}
 
 KOKORO_URL   = _cfg.get("tts_url", "http://localhost:8085/v1/audio/speech")
-KOKORO_MODEL = _cfg.get("kokoro_model", "kokoro")
+# Read "tts_model" (canonical key in config.json), fall back to legacy "kokoro_model"
+KOKORO_MODEL = _cfg.get("tts_model", _cfg.get("kokoro_model", "mlx-community/Kokoro-82M-bf16"))
 TTS_VOICE    = _cfg.get("tts_voice", "af_bella")
 
 _WRITE_VERBS = (
@@ -54,16 +55,21 @@ def run(task, app="", ctx=""):
     try:
         resp = requests.post(
             KOKORO_URL,
-            json={"model": KOKORO_MODEL, "input": clean, "voice": TTS_VOICE},
+            json={
+                "model": KOKORO_MODEL,
+                "input": clean,
+                "voice": TTS_VOICE,
+                "response_format": "wav",
+            },
             stream=True,
-            timeout=20,
+            timeout=30,
         )
         if resp.status_code != 200:
             # Fallback to macOS say
             subprocess.Popen(["say", clean])
             return f"🔊 (fallback) Speaking: {clean}"
 
-        tmp = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
+        tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
         for chunk in resp.iter_content(chunk_size=4096):
             tmp.write(chunk)
         tmp.close()