feat: VoiceCoach v2.1 - GPU local, Ollama LLM, settings panel, token server separation

- pipeline.py: switched LLM from OpenClaw/Claude to local Ollama, added settings loading, sentence aggregator, async SpeechBrain, tuned VAD
- main.py: simplified (no more token server thread, just pipeline)
- token_server.py: NEW - standalone token + settings API server
- docker-compose.yml: all services on GPU machine, added token-server service, CUDA speaches, Ollama LLM
- web-client/index.html: added settings panel with sliders
- web-client/app.js: settings management, localStorage persistence, push to server
- web-client/style.css: settings panel styles
- web-client/nginx.conf: proxy to token-server instead of pipecat-agent
- patch_livekit.py: updated with proper indentation-safe patching

pipecat-agent/patch_livekit.py

@@ -0,0 +1,58 @@
+"""Patch pipecat LiveKit transport bugs at build time."""
+import pipecat.transports.services.livekit as lk_mod
+
+lk_path = lk_mod.__file__
+print(f"Patching {lk_path}")
+content = open(lk_path).read()
+
+# FIX 1: Remove direct push_frame(pipecat_audio_frame) that bypasses VAD
+old1 = "await self.push_frame(pipecat_audio_frame)"
+if old1 in content:
+    content = content.replace(old1, "pass  # PATCHED: removed VAD bypass")
+    print("FIX 1: Removed push_frame(pipecat_audio_frame) VAD bypass")
+
+# FIX 2: Wrap the on_connected call in connect() method with try/except
+# The line is: "            await self._callbacks.on_connected()" (12 spaces)
+# We replace it with a try/except at the same indentation level
+old2 = "            await self._callbacks.on_connected()\n"
+new2 = (
+    "            try:\n"
+    "                await self._callbacks.on_connected()\n"
+    "            except Exception:\n"
+    "                pass  # PATCHED: suppress duplicate callback\n"
+)
+if old2 in content:
+    content = content.replace(old2, new2)
+    print("FIX 2: Wrapped on_connected with try/except")
+
+open(lk_path, "w").write(content)
+
+# Verify syntax
+import py_compile
+try:
+    py_compile.compile(lk_path, doraise=True)
+    print("Syntax check: PASS")
+except py_compile.PyCompileError as e:
+    print(f"Syntax check: FAIL - {e}")
+    # Show the problematic area
+    lines = open(lk_path).readlines()
+    for i in range(130, min(145, len(lines))):
+        print(f"  {i+1}: {lines[i]}", end="")
+
+# FIX 3: Guard set_result on already-resolved Future in livekit SDK
+try:
+    import livekit.rtc.room as room_mod
+    room_path = room_mod.__file__
+    room_content = open(room_path).read()
+    if ("self._first_sid_future.set_result" in room_content and
+            ".done()" not in room_content.split("_first_sid_future")[1][:200]):
+        room_content = room_content.replace(
+            "self._first_sid_future.set_result",
+            "not self._first_sid_future.done() and self._first_sid_future.set_result"
+        )
+        open(room_path, "w").write(room_content)
+        print(f"FIX 3: Patched {room_path}")
+except Exception as e:
+    print(f"FIX 3: Skipped: {e}")
+
+print("All patches complete")