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_harmony.py

@@ -0,0 +1,55 @@
+from llmux.harmony import extract_final_text, HarmonyStreamFilter
+
+
+def test_extract_llamacpp_format():
+    text = '<|channel|>analysis<|message|>User greeting. Simple.<|end|><|start|>assistant<|channel|>final<|message|>Hello! How can I help you today?'
+    assert extract_final_text(text) == "Hello! How can I help you today?"
+
+
+def test_extract_llamacpp_with_end_tag():
+    text = '<|channel|>analysis<|message|>thinking...<|end|><|start|>assistant<|channel|>final<|message|>The answer is 42.<|end|>'
+    assert extract_final_text(text) == "The answer is 42."
+
+
+def test_extract_transformers_format():
+    text = 'analysisUser greeting. Just respond friendly.assistantfinalHello! I am doing great.'
+    assert extract_final_text(text) == "Hello! I am doing great."
+
+
+def test_extract_non_harmony_passthrough():
+    text = "Hello! I'm doing well, thanks for asking."
+    assert extract_final_text(text) == text
+
+
+def test_stream_filter_llamacpp():
+    f = HarmonyStreamFilter()
+    chunks = [
+        "<|channel|>", "analysis", "<|message|>", "User ", "greeting.",
+        "<|end|>", "<|start|>", "assistant", "<|channel|>", "final",
+        "<|message|>", "Hello!", " How ", "are you?"
+    ]
+    output = ""
+    for c in chunks:
+        output += f.feed(c)
+    output += f.flush()
+    assert output == "Hello! How are you?"
+
+
+def test_stream_filter_transformers():
+    f = HarmonyStreamFilter()
+    chunks = ["analysis", "User ", "greeting.", "assistant", "final", "Hello!", " Great day!"]
+    output = ""
+    for c in chunks:
+        output += f.feed(c)
+    output += f.flush()
+    assert output == "Hello! Great day!"
+
+
+def test_stream_filter_non_harmony():
+    f = HarmonyStreamFilter()
+    chunks = ["Hello", " world", "!"]
+    output = ""
+    for c in chunks:
+        output += f.feed(c)
+    output += f.flush()
+    assert output == "Hello world!"