Remove all auth to fix GPU connection

src/inference/qwen.rs

@@ -6,15 +6,10 @@ use serde_json::Value;
 pub async fn generate(redacted_prompt: &str) -> Result<(String, String, String)> {
     let vllm_url = std::env::var("VLLM_URL")
         .unwrap_or_else(|_| "http://localhost:8000/v1/completions".to_string());
-    let api_key = std::env::var("VLLM_API_KEY")
-        .unwrap_or_else(|_| "abc-123".to_string());
-
-    // Append API key as query parameter (ngrok-friendly)
-    let url_with_key = format!("{}?api_key={}", vllm_url, api_key);
 
     let client = reqwest::Client::new();
     let resp = client
-        .post(&url_with_key)
+        .post(&vllm_url)
         .header("ngrok-skip-browser-warning", "true")
         .json(&serde_json::json!({
             "model": "Qwen/Qwen2.5-7B-Instruct",