fix: Open WebUI integration — Harmony stripping, VRAM eviction, concurrency lock

- Add harmony.py: strip GPT-OSS-20B analysis/thinking channel from both
  streaming and non-streaming responses (HarmonyStreamFilter + extract_final_text)
- Add per-model asyncio.Lock in llamacpp backend to prevent concurrent C++
  access that caused container segfaults (exit 139)
- Fix chat handler swap for streaming: move inside _stream_generate within
  lock scope (was broken by try/finally running before stream was consumed)
- Filter /v1/models to return only LLM models (hide ASR/TTS from chat dropdown)
- Correct Qwen3.5-4B estimated_vram_gb: 4 → 9 (actual allocation ~8GB)
- Add GPU memory verification after eviction with retry loop in vram_manager
- Add HF_TOKEN_PATH support in main.py for gated model access
- Add /v1/audio/models and /v1/audio/voices discovery endpoints (no auth)
- Add OOM error handling in both backends and chat route
- Add AUDIO_STT_SUPPORTED_CONTENT_TYPES for webm/wav/mp3/ogg
- Add performance test script (scripts/perf_test.py)
- Update tests to match current config (42 tests pass)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

kischdle/llmux/tests/test_config.py

@@ -13,10 +13,10 @@ def test_physical_model_has_required_fields():
     physical, _ = load_models_config()
     qwen = physical["qwen3.5-9b-fp8"]
     assert qwen.type == "llm"
-    assert qwen.backend == "transformers"
-    assert qwen.model_id == "lovedheart/Qwen3.5-9B-FP8"
-    assert qwen.estimated_vram_gb == 9
-    assert qwen.supports_vision is True
+    assert qwen.backend == "llamacpp"
+    assert qwen.model_id == "unsloth/Qwen3.5-9B-GGUF"
+    assert qwen.estimated_vram_gb == 10
+    assert qwen.supports_vision is False
     assert qwen.supports_tools is True