perf: eliminate loop string allocations and proxy mem spikes

mcp-stdio/src/lib.rs

@@ -6,8 +6,9 @@ pub async fn read_mcp_message<R: tokio::io::AsyncRead + Unpin>(
     stdin: &mut BufReader<R>,
 ) -> Option<String> {
     let mut length = 0;
+    let mut line = String::new();
     loop {
-        let mut line = String::new();
+        line.clear();
         if stdin.read_line(&mut line).await.unwrap_or(0) == 0 {
             return None;
         }