feat: Add Dify audio transcription with VAD chunking and SSE progress

- Add audio file upload transcription via Dify STT API
- Implement VAD-based audio segmentation in sidecar (3-min chunks)
- Add SSE endpoint for real-time transcription progress updates
- Fix chunk size enforcement for reliable uploads
- Add retry logic with exponential backoff for API calls
- Support Python 3.13+ with audioop-lts package
- Update frontend with Chinese progress messages and chunk display
- Improve start.sh health check with retry loop

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

start.sh

@@ -173,9 +173,23 @@ start_backend() {
     local backend_pid=$!
     echo "BACKEND_PID=$backend_pid" >> "$PID_FILE"
 
-    # 等待啟動
-    sleep 2
+    # 等待啟動（最多等待 15 秒）
+    local max_wait=15
+    local waited=0
+    log_info "等待後端服務啟動..."
 
+    while [ $waited -lt $max_wait ]; do
+        sleep 1
+        waited=$((waited + 1))
+
+        # 檢查健康狀態
+        if curl -s http://localhost:$BACKEND_PORT/api/health > /dev/null 2>&1; then
+            log_success "後端服務已啟動 (PID: $backend_pid, Port: $BACKEND_PORT)"
+            return 0
+        fi
+    done
+
+    # 最後再檢查一次 port 狀態
     if check_port $BACKEND_PORT; then
         log_success "後端服務已啟動 (PID: $backend_pid, Port: $BACKEND_PORT)"
     else