move kokoro to separate deployment

apps/llama/configs/config.yaml

@@ -20,7 +20,6 @@ hooks:
     preload:
       - "Qwen3.5-0.8B-GGUF-nothink:Q4_K_XL"
       - "whisper-small"
-      - "outetts-tts"
 
 # matrix replaces groups (they are mutually exclusive).
 # The small 0.8B model runs alongside any LLM.
@@ -29,7 +28,6 @@ matrix:
   vars:
     q8: "Qwen3.5-0.8B-GGUF-nothink:Q4_K_XL"
     stt: "whisper-small"
-    tts: "outetts-tts"
     flux: "flux2-klein-4b:Q4_K_M"
     coder: "Qwen3-Coder-Next-GGUF:Q4_K_M"
     q35t: "Qwen3.5-35B-A3B-GGUF:Q4_K_M"
@@ -56,9 +54,9 @@ matrix:
 
   sets:
     # any LLM can run alongside the small always-on model + STT + TTS (all CPU, no VRAM cost)
-    with_q8: "(coder | q35t | q35nt | q35ht | q35hnt | q4t | q4nt | q4ht | q4hnt | g26xl | g26q2 | ge4xl | ge2xl | q36t | q36nt | haut | haunt | mtpt | mtpnt) & q8 & stt & tts"
-    # FLUX runs alone — evicts everything including q8, but keeps STT+TTS for voice during image gen
-    image_gen: "flux & stt & tts"
+    with_q8: "(coder | q35t | q35nt | q35ht | q35hnt | q4t | q4nt | q4ht | q4hnt | g26xl | g26q2 | ge4xl | ge2xl | q36t | q36nt | haut | haunt | mtpt | mtpnt) & q8 & stt"
+    # FLUX runs alone — evicts everything including q8, but keeps STT for voice during image gen
+    image_gen: "flux & stt"
 
 models:
   "Qwen3-Coder-Next-GGUF:Q4_K_M":
@@ -252,19 +250,6 @@ models:
         --threads 6
         --no-gpu
 
-  # TTS via OuteTTS 0.3 1B + WavTokenizer vocoder (CPU-only, always loaded)
-  # Models auto-downloaded from HuggingFace on first start
-  # OuteTTS 0.3 1B: ~1GB RAM, WavTokenizer: ~600MB RAM
-  # Exposes /v1/audio/speech compatible with OpenAI TTS API
-  "outetts-tts":
-    checkEndpoint: none
-    cmd: |
-      llama-server
-        -hf OuteAI/OuteTTS-0.3-1B-GGUF
-        -hff OuteTTS-0.3-1B-Q8_0.gguf
-        -mv /root/.cache/huggingface/hub/models--ggml-org--WavTokenizer/snapshots/0c97fdc098158ec9bf4e703cd5f81a5aa20520e6/WavTokenizer-Large-75-F16.gguf
-        -c 4096
-        ${cpu_args}
 
   # Image generation via stable-diffusion.cpp (sd-server)
   # Models must be pre-downloaded to /root/.cache/sd/