refactor: split openai/ into ollama/, openwebui/, searxng/

Move services out of the monolithic openai/docker-compose.yml:
- ollama/ — ollama GPU + CPU inference servers
- openwebui/ — open-webui chat UI (uses env var for ANTHROPIC_API_KEY)
- searxng/ — SearXNG container + searxng-mcp MCP server (port 3102)

openai/ now contains only: litellm, langfuse, qdrant, faster-whisper,
silero-tts, pipecat.

searxng-mcp exposes a single searxng_search tool via MCP HTTP on :3102.
Registered in ~/.claude.json as the "searxng" MCP server.

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

ollama/docker-compose.yml

@@ -0,0 +1,27 @@
+services:
+  ollama:
+    image: ollama/ollama
+    container_name: ollama
+    ports:
+      - "11436:11434"
+    volumes:
+      - /mnt/ssd/ai/ollama:/root/.ollama
+    restart: always
+    environment:
+      # Allow qwen3:8b + qwen2.5:1.5b to coexist in VRAM (~6.7-7.7 GB on 8 GB GPU)
+      - OLLAMA_MAX_LOADED_MODELS=2
+      # One GPU inference at a time — prevents compute contention between models
+      - OLLAMA_NUM_PARALLEL=1
+      # Force all layers to GPU — fail instead of falling back to CPU
+      - OLLAMA_NUM_GPU=999
+    runtime: nvidia
+    mem_limit: 4g
+
+  ollama-cpu:
+    image: ollama/ollama
+    container_name: ollama-cpu
+    ports:
+      - "11435:11434"
+    volumes:
+      - /mnt/ssd/ai/ollama-cpu:/root/.ollama
+    restart: always